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>
When a process exec()'s, all the maps are retired, but we keep the hist
entries around which hold references to those outdated maps.
If the same library gets mapped in for which we have hist entries, a new
map will be created. But when we take a perf entry hit within that map,
we'll find the existing hist entry with the older map.
This causes symbol translations to be done incorrectly. For example,
the perf entry processing will lookup the correct uptodate map entry and
use that to calculate the symbol and DSO relative address. But later
when we update the histogram we'll translate the address using the
outdated map file instead leading to conditions such as out-of-range
offsets in symbol__inc_addr_samples().
Therefore, update the map of the hist_entry dynamically at lookup/
creation time.
Signed-off-by: David S. Miller <davem@davemloft.net>
Cc: stable@kernel.org
Link: http://lkml.kernel.org/r/20120327.031418.1220315351537060808.davem@davemloft.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The callchain stdio mode display was written using a sorted by symbol
report. In this mode we have only one callchain root per hist so we
forgot to handle cases where we have multiple callchain root, as in per
dso sorting for example.
Fix this by handling these roots like any other branch, with the hist as
the parent.
Before:
1.97% libpthread-2.12.1.so
|
--- __libc_write
create_worker
bench_sched_messaging
cmd_bench
run_builtin
main
__libc_start_main
|
--- __libc_read
create_worker
bench_sched_messaging
cmd_bench
run_builtin
main
__libc_start_main
After:
1.97% libpthread-2.12.1.so
|
|--36.97%-- __libc_write
| create_worker
| bench_sched_messaging
| cmd_bench
| run_builtin
| main
| __libc_start_main
|
|--31.47%-- __libc_read
| create_worker
| bench_sched_messaging
| cmd_bench
| run_builtin
| main
| __libc_start_main
...
Single roots keep their entry without percentage because they have
the same overhead than the hist they refer to. ie: 100% in fractal
mode and the percentage of the hist in graph mode:
0.00% [k] reschedule_interrupt
|
--- default_idle
amd_e400_idle
cpu_idle
start_secondary
Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1332526010-15400-1-git-send-email-fweisbec@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf diff command is broken since:
perf hists: Threaded addition and sorting of entries
commit 1980c2ebd7
Several places were broken:
- hists data need to be collected into opened sessions instead
of into events
- session's hists data need to be initialized properly when the
session is created
- hist_entry__pcnt_snprintf: the percentage and displacement
buffer preparation must not use 'ret' because it's used
as a pointer to the final buffer
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/20120322133726.GB1601@m.brq.redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Recently we made perf_evsel__init call hists__init, which broke the perf
python binding:
[root@emilia linux]# ./tools/perf/python/twatch.py
Traceback (most recent call last):
File "./tools/perf/python/twatch.py", line 16, in <module>
import perf
ImportError: /home/acme/git/build/perf/python/perf.so: undefined symbol: hists__init
Fix it by moving the hists__init function to its only caller, evsel.c.
This way we avoid dragging in other parts of tools/perf/util/ to the
perf python binding.
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
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-5nffmdt5mu6ozxgj54oi4qon@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We were doing parts of it in hists__collapse_resort and parts of it in
hists__output_resort, leading to a bogus total_period.
Fix it by doing just the filtering operation when collapsing because
there we know that the Zoom operations adds filters just what is in
hists->entries, not to the new batch of entries being collapsed.
And move all the nr_entries + total_period recalculation to
hists__output_resort since we will traverse all entries anyway there.
Problem introduced when developing threaded addition of new batches
of hist_entries, i.e. post v3.1.
Reported-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
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-8xyh165h7hmwy0696hu25en6@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Following the 'perf report' model we don't zap hist_entry instances from
the rb tree, we just keep them with he->filtered set to a mask of the
filters applied to it (thread, parent, DSO so far).
In top we need to decay even filtered entries, but we better not touch
total_period for them...
Now everything seems to work when filters are applied on top as they
worked in 'report', i.e. both dynamic and static hist entry browsing
works with filters.
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
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-yt4xsbq20u9x9ypuwwyw2kao@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf report -n option was broken because it was not reporting the
correct number of samples depending on the sorting mode. By default,
samples are sorted by comm,dso,sym. That means that samples for the same
command (binary) get collapsed.
The hists__collapse_insert_entry() had a bug whereby it was aggregating
the number of events observed (periods) but not the number of samples.
Consequently, the number of samples reported could be below reality. The
percentage remained correct because based on the periods.
This patch fixes the problem by also aggregating the number of samples.
Here is an example:
$ perf report -n --stdio
12.38% 842 pong [kernel.kallsyms] [k] __lock_acquire
Here pong (a ctxsw stress test), is the only program running
and thus it is the only one responsible for the lock_acquire samples.
If we change the sorting mode:
$ perf report -n --stdio --sort=sym
12.38% 1732 [k] __lock_acquire
The actual number of samples is shown.
With the fix:
$ perf report -n --stdio
12.38% 1732 pong [kernel.kallsyms] [k] __lock_acquire
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20111003093815.GA6393@quad
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This actually fixes several problems we had in the old 'perf top':
1. Unresolved symbols not show, limitation that came from the old
"KernelTop" codebase, to solve it we would need to do changes
that would make sym_entry have most of the hist_entry fields.
2. It was using the number of samples, not the sum of sample->period.
And brings the --sort code that allows us to have all the views in
'perf report', for instance:
[root@emilia ~]# perf top --sort dso
PerfTop: 5903 irqs/sec kernel:77.5% exact: 0.0% [1000Hz cycles], (all, 8 CPUs)
------------------------------------------------------------------------------
31.59% libcrypto.so.1.0.0
21.55% [kernel]
18.57% libpython2.6.so.1.0
7.04% libc-2.12.so
6.99% _backend_agg.so
4.72% sshd
1.48% multiarray.so
1.39% libfreetype.so.6.3.22
1.37% perf
0.71% libgobject-2.0.so.0.2200.5
0.53% [tg3]
0.48% libglib-2.0.so.0.2200.5
0.44% libstdc++.so.6.0.13
0.40% libcairo.so.2.10800.8
0.38% libm-2.12.so
0.34% umath.so
0.30% libgdk-x11-2.0.so.0.1800.9
0.22% libpthread-2.12.so
0.20% libgtk-x11-2.0.so.0.1800.9
0.20% librt-2.12.so
0.15% _path.so
0.13% libpango-1.0.so.0.2800.1
0.11% libatlas.so.3.0
0.09% ft2font.so
0.09% libpangoft2-1.0.so.0.2800.1
0.08% libX11.so.6.3.0
0.07% [vdso]
0.06% cyclictest
^C
All the filter lists can be used as well: --dsos, --comms, --symbols,
etc.
The 'perf report' TUI is also reused, being possible to apply all the
zoom operations, do annotation, etc.
This change will allow multiple simplifications in the symbol system as
well, that will be detailed in upcoming changesets.
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
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-xzaaldxq7zhqrrxdxjifk1mh@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
By using a mutex just for inserting and rotating two hist_entry rb
trees, so that when sorting we can get the last batch of entries created
from the ring buffer, merge it with whatever we have processed so far
and show the output while new entries are being added.
The 'report' tool continues, for now, to do it without threading, but
will use this in the future to allow visualization of results in long
perf.data sessions while the entries are being processed.
The new 'top' tool will be the first user.
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
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-9b05atsn0q6m7fqgrug8fk2i@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
