First clarifying that this kind of binding is not a replacement or an
equivalent to the 'perf script' way of using python with perf.
The 'perf script' way is to process events and look at a given script
for some python function that matches the events to pass each event for
processing.
This is a python module, i.e. everything is driven from the python
script, that merely uses "import perf" or "from perf import".
perf script is focused on tracepoints, this binding is focused on profiling as
an initial target. More work is needed to make available tracepoint specific
variables as event variables accessible via this binding.
There is one example of such usage model, in
tools/perf/python/twatch.py, a tool to watch "cycles" events together
with task (fork, exit) and comm perf events.
For now, due to me not being able to grok how python distutils cope with
building C extensions outside the sources dir the install target just
builds it, I'm using it as:
[root@emilia linux]# export PYTHONPATH=~acme/git/build/perf/lib.linux-x86_64-2.6/
[root@emilia linux]# tools/perf/python/twatch.py
cpu: 4, pid: 30126, tid: 30126 { type: mmap, pid: 30126, tid: 30126, start: 0x4, length: 0x82e9ca03, offset: 0, filename: }
cpu: 6, pid: 47, tid: 47 { type: mmap, pid: 47, tid: 47, start: 0x6, length: 0xbef87c36, offset: 0, filename: }
cpu: 1, pid: 0, tid: 0 { type: mmap, pid: 0, tid: 0, start: 0x1, length: 0x775d1904, offset: 0, filename: }
cpu: 7, pid: 0, tid: 0 { type: mmap, pid: 0, tid: 0, start: 0x7, length: 0xc750aeb6, offset: 0, filename: }
cpu: 5, pid: 2255, tid: 2255 { type: mmap, pid: 2255, tid: 2255, start: 0x5, length: 0x76669635, offset: 0, filename: }
cpu: 0, pid: 0, tid: 0 { type: mmap, pid: 0, tid: 0, start: 0, length: 0x6422ef6b, offset: 0, filename: }
cpu: 2, pid: 2255, tid: 2255 { type: mmap, pid: 2255, tid: 2255, start: 0x2, length: 0xe078757a, offset: 0, filename: }
cpu: 1, pid: 5769, tid: 5769 { type: fork, pid: 30127, ppid: 5769, tid: 30127, ptid: 5769, time: 103893991270534}
cpu: 6, pid: 30127, tid: 30127 { type: comm, pid: 30127, tid: 30127, comm: ls }
cpu: 6, pid: 30127, tid: 30127 { type: exit, pid: 30127, ppid: 30127, tid: 30127, ptid: 30127, time: 103893993273024}
The first 8 mmap events in this 8 way machine are a mistery that is still being
investigated.
More of the tools/perf/util/ APIs will be exposed via this python binding as
the need arises. For now the focus is on creating events and processing them,
symbol resolution is an obvious next step, with tracepoint variables as a close
second step.
Cc: Clark Williams <williams@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
I.e. stash the overwrite mode in struct perf_evlist and act accordingly
in perf_evlist__read_on_cpu, not checking for overwrites and touching
the tail after consuming one event, like perf record does, for instance.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Right now this function is only used by perf top, that uses PROT_READ
only, i.e. overwrite mode, so no PERF_RECORD_LOST events are generated,
but don't forget those events.
The patch that moved this out of perf top was made so that this routine
could be used by 'perf probe' in the uprobes patchset, so perhaps there
they need to check for LOST events and warn the user, as will be done in
the following patches that will switch 'perf top' to non overwrite mode
(mmap with PROT_READ|PROT_WRITE).
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add filters support for available variable list.
Default filter is "!__k???tab_*&!__crc_*" for filtering out
automatically generated symbols.
The format of filter rule is "[!]GLOBPATTERN", so you can use wild
cards. If the filter rule starts with '!', matched variables are filter
out.
e.g.:
# perf probe -V schedule --externs --filter=cpu*
Available variables at schedule
@<schedule+0>
cpumask_var_t cpu_callout_mask
cpumask_var_t cpu_core_map
cpumask_var_t cpu_isolated_map
cpumask_var_t cpu_sibling_map
int cpu_number
long unsigned int* cpu_bit_bitmap
...
Cc: 2nddept-manager@sdl.hitachi.co.jp
Cc: Chase Douglas <chase.douglas@canonical.com>
Cc: Franck Bui-Huu <fbuihuu@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20110120141539.25915.43401.stgit@ltc236.sdl.hitachi.co.jp>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
[ committer note: Removed the elf.h include as it was fixed up in e80711c]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add strfilter for general purpose string filter.
Every filter rules are descrived by glob matching pattern and '!' prefix
which means Logical NOT.
A strfilter consists of those filter rules connected with '&' and '|'.
A set of rules can be folded by using '(' and ')'.
It also accepts spaces around rules and those operators.
Format:
<rule> ::= <glob-exp> | "!" <rule> | <rule> <op> <rule> | "(" <rule> ")"
<op> ::= "&" | "|"
e.g.:
"(add* | del*) & *timer" filter rules pass strings which start with add
or del and end with timer.
This will be used by perf probe --filter.
Changes in V2:
- Fix to check result of strdup() and strfilter__alloc().
- Encapsulate and simplify interfaces as like regex(3).
Cc: 2nddept-manager@sdl.hitachi.co.jp
Cc: Franck Bui-Huu <fbuihuu@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20110120141530.25915.12673.stgit@ltc236.sdl.hitachi.co.jp>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Out of the {con,des}structor, as in interpreted language bindings we will
need to go back from the wrapper object to the real thing. In that case
using container_of will save us to have an extra pointer in the perf_evsel
struct.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To avoid linking more stuff in the python binding I'm working on, future
csets will make the sample type be taken from the evsel itself, but for
that we need to first have one file per cpu and per sample_type, not a
single perf.data file.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To untangle it from struct thread handling, that is tied to symbols, etc.
Right now in the python bindings I'm working on I need just a subset of
the util/ files, untangling it allows me to do that.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It is needed because it will call parse_event for each tracepoint
name that matches, and we pass the perf_evlist via opt->value.
Problem introduced in 4503fdd where my assumption about opt being
always non NULL made me not look at callers of parse_events outside
builtin-*.c.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This patch gives the ability to 'perf record' to detect when its stdout
has been redirected to a pipe. There's now no more need to add '-o -'
switch in this case.
However '-o <path>' option has always precedence, that is if specified
and stdout has been connected via a pipe then the output will go into
the specified output.
LKML-Reference: <m3ipxo966i.fsf@gmail.com>
Signed-off-by: Franck Bui-Huu <fbuihuu@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The callchains are fed with an array of a fixed size.
As a result we iterate over each callchains three times:
- 1st to resolve symbols
- 2nd to filter out context boundaries
- 3rd for the insertion into the tree
This also involves some pairs of memory allocation/deallocation
everytime we insert a callchain, for the filtered out array of
addresses and for the array of symbols that comes along.
Instead, feed the callchains through a linked list with persistent
allocations. It brings several pros like:
- Merge the 1st and 2nd iterations in one. That was possible before
but in a way that would involve allocating an array slightly taller
than necessary because we don't know in advance the number of context
boundaries to filter out.
- Much lesser allocations/deallocations. The linked list keeps
persistent empty entries for the next usages and is extendable at
will.
- Makes it easier for multiple sources of callchains to feed a
stacktrace together. This is deemed to pave the way for cfi based
callchains wherein traditional frame pointer based kernel
stacktraces will precede cfi based user ones, producing an overall
callchain which size is hardly predictable. This requirement
makes the static array obsolete and makes a linked list based
iterator a much more flexible fit.
Basic testing on a big perf file containing callchains (~ 176 MB)
has shown a throughput gain of about 11% with perf report.
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This test will generate random numbers of calls to some getpid syscalls,
then establish an mmap for a group of events that are created to monitor
these syscalls.
It will receive the events, using mmap, use its PERF_SAMPLE_ID generated
sample.id field to map back to its respective perf_evsel instance.
Then it checks if the number of syscalls reported as perf events by the
kernel corresponds to the number of syscalls made.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
It looks like we should check if cpus is NULL after
cpus = cpu_map__new(NULL);
in test__open_syscall_event_on_all_cpus().
LKML-Reference: <20110114230050.GA7011@localhost>
Signed-off-by: Han Pingtian <phan@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Adopting the new model used in 'perf record', where we don't have a map
per thread per cpu, instead we have an mmap per cpu, established on the
first fd for that cpu and ask the kernel using the
PERF_EVENT_IOC_SET_OUTPUT ioctl to send events for the other fds on that
cpu for the one with the mmap.
The methods moved from perf_evsel to perf_evlist, but for easing review
they were modified in place, in evsel.c, the next patch will move the
migrated methods to evlist.c.
With this 'perf top' now uses the same mmap model used by 'perf record'
and the next patches will make 'perf record' use these new routines,
establishing a common codebase for both tools.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Now that it handles group_fd and inherit we can use it, sharing it with
stat.
Next step: 'perf record' should use, then move the mmap_array out of
->priv and into perf_evsel, with top and record sharing this, and at the
same time, write a 'perf test' stress test.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf_evsel__open now have an extra boolean argument specifying if
event grouping is desired.
The first file descriptor created on a CPU becomes the group leader.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Killing two more perf wide global variables: nr_counters and evsel_list
as a list_head.
There are more operations that will need more fields in perf_evlist,
like the pollfd for polling all the fds in a list of evsel instances.
Use option->value to pass the evsel_list to parse_{events,filters}.
LKML-Reference: <new-submission>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The -Wstack-protector and -Wvolatile-register-var warnings, for
instance, are not supported by gcc 3.4.6.
So fix by doing the same check we already do for -fstack-protector-all.
With this and the other patches in this series, perf builds unmodified
on, for instance, RHEL4.
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
[acme@localhost linux]$ make O=~acme/git/build/perf -C tools/perf
make: Entering directory `/home/acme/git/linux/tools/perf'
Makefile:526: No libdw.h found or old libdw.h found or elfutils is older than 0.138, disables dwarf support. Please install new elfutils-devel/libdw-dev
Makefile:582: newt not found, disables TUI support. Please install newt-devel or libnewt-dev
CC /home/acme/git/build/perf/builtin-annotate.o
In file included from builtin-annotate.c:23:
util/parse-events.h:26: warning: declaration of 'evsel_list' shadows a global declaration
util/parse-events.h:12: warning: shadowed declaration is here
make: *** [/home/acme/git/build/perf/builtin-annotate.o] Error 1
make: Leaving directory `/home/acme/git/linux/tools/perf'
[acme@localhost linux]$ gcc --version | head -1
gcc (GCC) 3.4.6 20060404 (Red Hat 3.4.6-11)
[acme@localhost linux]$
Fix it by renaming the parameter to evlist.
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We need the definiton for __always_inline in bitops.h to fix the build
on distros where it isn't available or compiler.h doesn't get included
indirectly.
One of the fixes needed to build perf on RHEL4 systems, for instance.
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When some of CPUs are offline:
# cat /sys/devices/system/cpu/online
0,6-31
perf test will fail on #3 testcase:
3: detect open syscall event on all cpus:
--- start ---
perf_evsel__read_on_cpu: expected to intercept 111 calls on cpu 0, got 681
perf_evsel__read_on_cpu: expected to intercept 112 calls on cpu 1, got 117
perf_evsel__read_on_cpu: expected to intercept 113 calls on cpu 2, got 118
perf_evsel__read_on_cpu: expected to intercept 114 calls on cpu 3, got 119
perf_evsel__read_on_cpu: expected to intercept 115 calls on cpu 4, got 120
perf_evsel__read_on_cpu: expected to intercept 116 calls on cpu 5, got 121
perf_evsel__read_on_cpu: expected to intercept 117 calls on cpu 6, got 122
perf_evsel__read_on_cpu: expected to intercept 118 calls on cpu 7, got 123
perf_evsel__read_on_cpu: expected to intercept 119 calls on cpu 8, got 124
perf_evsel__read_on_cpu: expected to intercept 120 calls on cpu 9, got 125
perf_evsel__read_on_cpu: expected to intercept 121 calls on cpu 10, got 126
....
This patch try to use 'cpus->map[cpu]' when setting cpu affinity, and
will check the return code of sched_setaffinity()
LKML-Reference: <20110120114707.GA11781@hpt.nay.redhat.com>
Signed-off-by: Han Pingtian <phan@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
In ARM's Thumb mode the bottom bit of the symbol address is set to mark
the function as Thumb; the instructions are in reality 2 or 4 byte on 2
byte alignments, and when the +1 address is used in annotate it causes
objdump to disassemble invalid instructions.
The patch removes that bottom bit during symbol loading.
Many thinks to Dave Martin for comments on an initial version of the
patch.
(For reference this corresponds to this bug
https://bugs.launchpad.net/linux-linaro/+bug/677547 )
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Dave Martin <dave.martin@linaro.org>
LKML-Reference: <20110121163922.GA31398@davesworkthinkpad>
Signed-off-by: Dr. David Alan Gilbert <david.gilbert@linaro.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
perf tools: Fix tracepoint id to string perf.data header table
perf tools: Fix handling of wildcards in tracepoint event selectors
powerpc: perf: Fix frequency calculation for overflowing counters
