This commit adds support for a new modifier "D", which requests that the
event, or group of events, be pinned to the PMU.
The "p" modifier is already taken for precise, and "P" may be used in
future to mean "fully precise".
So we use "D", which stands for pinneD - and looks like a padlock, or if
you're using the ":D" syntax perf smiles at you.
This is an oft-requested feature from our HW folks, who want to be able
to run a large number of events, but also want 100% accurate results for
instructions per cycle.
Comparison of results with and without pinning:
$ perf stat -e '{cycles,instructions}:D' -e cycles,instructions,...
79,590,480,683 cycles # 0.000 GHz
166,123,716,524 instructions # 2.09 insns per cycle
# 0.11 stalled cycles per insn
79,352,134,463 cycles # 0.000 GHz [11.11%]
165,178,301,818 instructions # 2.08 insns per cycle
# 0.11 stalled cycles per insn [11.13%]
As you can see although perf does a very good job of scaling the values
in the non-pinned case, there is some small discrepancy.
The patch is fairly straight forward, the one detail is that we need to
make sure we only request pinning for the group leader when we have a
group.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Tested-by: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1375795686-4226-1-git-send-email-michael@ellerman.id.au
[ Use perf_evsel__is_group_leader instead of open coded equivalent, as
suggested by Jiri Olsa ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When measuring workloads the startup phase -- doing page faults, dynamic
linking, opening files -- is often very different from the rest of the
workload. Especially with smaller kernels and using counter
multiplexing this can give significant measurement errors.
Multiplexing assumes that the workload is mostly the same over longer
periods. But at startup there is typically some spike of activity which
is relatively short. If many groups are multiplexing the one group
seeing the spike, and which is then scaled up over the time to run all
groups, may see a significant error.
Also in general it's often not useful to measure the startup, because it
is so different from the rest.
One way around this is to use interval mode and discard the first
sample, but this can be awkward because interval mode doesn't support
intervals of less than 100ms, and also a useful interval is not
necessarily the same as a useful startup delay.
This patch adds a new --initial-delay / -D option to skip measuring for
the startup phase. The time can be specified in ms
Here's a simple example:
perf stat -e page-faults bash -c 'for i in $(seq 100000) ; do true ; done'
...
3,721 page-faults
...
If we just wait 20 ms the number of page faults is 1/3 less:
perf stat -D 20 -e page-faults bash -c 'for i in $(seq 100000) ; do true ; done'
...
2,823 page-faults
...
So we filtered out most of the startup noise from bash.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1375490473-1503-4-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Adding 'S' event/group modifier to specify that the event value/s are
read by PERF_SAMPLE_READ sample type processing, instead of the period
value offered by lower layers.
There's additional behaviour change for 'S' modifier being specified on
event group:
Currently all the events within a group makes samples. If user now
specifies 'S' within group modifier, only the leader will trigger
samples. The rest of events in the group will have sampling disabled.
And same as for single events, values of all events within the group
(including leader) are read by PERF_SAMPLE_READ sample type processing.
Following example will create event group with cycles and cache-misses
events, setting the cycles as group leader and the only event to
actually sample. Both cycles and cache-misses event period values are
read by PERF_SAMPLE_READ sample type processing with PERF_FORMAT_GROUP
read format.
Example:
$ perf record -e '{cycles,cache-misses}:S' ls
...
$ perf report --group --show-total-period --stdio
...
# Samples: 36 of event 'anon group { cycles, cache-misses }'
# Event count (approx.): 12585593
#
# Overhead Period Command Shared Object Symbol
# .............. .............. ....... ................. ..........................
#
19.92% 1.20% 2505936 31 ls [kernel.kallsyms] [k] mark_held_locks
13.74% 0.47% 1729327 12 ls [kernel.kallsyms] [k] sched_clock_local
13.64% 23.72% 1716147 612 ls ld-2.14.90.so [.] check_match.10805
13.12% 23.22% 1650778 599 ls libc-2.14.90.so [.] _nl_intern_locale_data
11.24% 29.19% 1414554 753 ls [kernel.kallsyms] [k] sched_clock_cpu
8.50% 0.35% 1070150 9 ls [kernel.kallsyms] [k] check_chain_key
...
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-iyoinu3axi11mymwnh2b7fxj@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
With programs with very large functions it can be useful to distinguish
the callgraph nodes on more than just function names. So for example if
you have multiple calls to the same function, it ends up being separate
nodes in the chain.
This patch adds a new key field to the callgraph options, that allows
comparing nodes on functions (as today, default) and addresses.
Longer term it would be nice to also handle src lines, but that would
need more changes and address is a reasonable proxy for it today.
I right now reference the global params, as there was no simple way to
register a params pointer.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/n/tip-0uskktybf0e7wrnoi5e9b9it@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
For example, in an application with an expensive function implemented
with deeply nested recursive calls, the default call-graph presentation
is dominated by the different callchains within that function. By
ignoring these callees, we can collect the callchains leading into the
function and compactly identify what to blame for expensive calls.
For example, in this report the callers of garbage_collect() are
scattered across the tree:
$ perf report -d ruby 2>- | grep -m10 ^[^#]*[a-z]
22.03% ruby [.] gc_mark
--- gc_mark
|--59.40%-- mark_keyvalue
| st_foreach
| gc_mark_children
| |--99.75%-- rb_gc_mark
| | rb_vm_mark
| | gc_mark_children
| | gc_marks
| | |--99.00%-- garbage_collect
If we ignore the callees of garbage_collect(), its callers are coalesced:
$ perf report --ignore-callees garbage_collect -d ruby 2>- | grep -m10 ^[^#]*[a-z]
72.92% ruby [.] garbage_collect
--- garbage_collect
vm_xmalloc
|--47.08%-- ruby_xmalloc
| st_insert2
| rb_hash_aset
| |--98.45%-- features_index_add
| | rb_provide_feature
| | rb_require_safe
| | vm_call_method
Signed-off-by: Greg Price <price@mit.edu>
Tested-by: Jiri Olsa <jolsa@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20130623031720.GW22203@biohazard-cafe.mit.edu
Link: http://lkml.kernel.org/r/20130708115746.GO22203@biohazard-cafe.mit.edu
Cc: Fengguang Wu <fengguang.wu@intel.com>
[ remove spaces at beginning of line, reported by Fengguang Wu ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
