Added a minimal exec tracepoint. Exec is an important major event
in the life of a task, like fork(), clone() or exit(), all of
which we already trace.
[ We also do scheduling re-balancing during exec() - so it's useful
from a scheduler instrumentation POV as well. ]
If you want to watch a task start up, when it gets exec'ed is a good place
to start. With the addition of this tracepoint, exec's can be monitored
and better picture of general system activity can be obtained. This
tracepoint will also enable better process life tracking, allowing you to
answer questions like "what process keeps starting up binary X?".
This tracepoint can also be useful in ftrace filtering and trigger
conditions: i.e. starting or stopping filtering when exec is called.
Signed-off-by: David Smith <dsmith@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/4F314D19.7030504@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Small fixes, also includes an important fix from Stephane for system
wide monitoring, problem introduced recently in perf/core, in the pid
list patches.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The following commit:
b52956c perf tools: Allow multiple threads or processes in record, stat, top
introduced a bug in the thread_map code which caused perf record -a to
not setup system-wide monitoring properly.
$ taskset -c 1 noploop 1000 &
$ perf record -a -C 1 sleep 10
$ perf report -D | tail -20
cycles stats:
TOTAL events: 4413
MMAP events: 4025
COMM events: 340
SAMPLE events: 48
Here I was expecting about 10,000 samples and not 48.
In system-wide mode, the PID passed to perf_event_open() must be -1 and
it was 0. That caused the kernel to setup a per-process event on PID:0.
Consequently, the number of samples captured does not correspond to the
requested measurement.
The following one-liner fixes the problem for me with or without -C.
I would also suggest to change the malloc() to something that matches
the struct definition. thread_map->map[] is declared as int map[] and
not pid_t map[]. If map[] can only contain pids, then change the struct
definition.
Acked-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20120221145424.GA6757@quad
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Adding support to filter function trace event via perf
interface. It is now possible to use filter interface
in the perf tool like:
perf record -e ftrace:function --filter="(ip == mm_*)" ls
The filter syntax is restricted to the the 'ip' field only,
and following operators are accepted '==' '!=' '||', ending
up with the filter strings like:
ip == f1[, ]f2 ... || ip != f3[, ]f4 ...
with comma ',' or space ' ' as a function separator. If the
space ' ' is used as a separator, the right side of the
assignment needs to be enclosed in double quotes '"', e.g.:
perf record -e ftrace:function --filter '(ip == do_execve,sys_*,ext*)' ls
perf record -e ftrace:function --filter '(ip == "do_execve,sys_*,ext*")' ls
perf record -e ftrace:function --filter '(ip == "do_execve sys_* ext*")' ls
The '==' operator adds trace filter with same effect as would
be added via set_ftrace_filter file.
The '!=' operator adds trace filter with same effect as would
be added via set_ftrace_notrace file.
The right side of the '!=', '==' operators is list of functions
or regexp. to be added to filter separated by space.
The '||' operator is used for connecting multiple filter definitions
together. It is possible to have more than one '==' and '!='
operators within one filter string.
Link: http://lkml.kernel.org/r/1329317514-8131-8-git-send-email-jolsa@redhat.com
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Adding FILTER_TRACE_FN event field type for function tracepoint
event, so it can be properly recognized within filtering code.
Currently all fields of ftrace subsystem events share the common
field type FILTER_OTHER. Since the function trace fields need
special care within the filtering code we need to recognize it
properly, hence adding the FILTER_TRACE_FN event type.
Adding filter parameter to the FTRACE_ENTRY macro, to specify the
filter field type for the event.
Link: http://lkml.kernel.org/r/1329317514-8131-7-git-send-email-jolsa@redhat.com
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Adding perf registration support for the ftrace function event,
so it is now possible to register it via perf interface.
The perf_event struct statically contains ftrace_ops as a handle
for function tracer. The function tracer is registered/unregistered
in open/close actions.
To be efficient, we enable/disable ftrace_ops each time the traced
process is scheduled in/out (via TRACE_REG_PERF_(ADD|DELL) handlers).
This way tracing is enabled only when the process is running.
Intentionally using this way instead of the event's hw state
PERF_HES_STOPPED, which would not disable the ftrace_ops.
It is now possible to use function trace within perf commands
like:
perf record -e ftrace:function ls
perf stat -e ftrace:function ls
Allowed only for root.
Link: http://lkml.kernel.org/r/1329317514-8131-6-git-send-email-jolsa@redhat.com
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Adding TRACE_REG_PERF_OPEN and TRACE_REG_PERF_CLOSE to differentiate
register/unregister from open/close actions.
The register/unregister actions are invoked for the first/last
tracepoint user when opening/closing the event.
The open/close actions are invoked for each tracepoint user when
opening/closing the event.
Link: http://lkml.kernel.org/r/1329317514-8131-3-git-send-email-jolsa@redhat.com
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Adding a way to temporarily enable/disable ftrace_ops. The change
follows the same way as 'global' ftrace_ops are done.
Introducing 2 global ftrace_ops - control_ops and ftrace_control_list
which take over all ftrace_ops registered with FTRACE_OPS_FL_CONTROL
flag. In addition new per cpu flag called 'disabled' is also added to
ftrace_ops to provide the control information for each cpu.
When ftrace_ops with FTRACE_OPS_FL_CONTROL is registered, it is
set as disabled for all cpus.
The ftrace_control_list contains all the registered 'control' ftrace_ops.
The control_ops provides function which iterates ftrace_control_list
and does the check for 'disabled' flag on current cpu.
Adding 3 inline functions:
ftrace_function_local_disable/ftrace_function_local_enable
- enable/disable the ftrace_ops on current cpu
ftrace_function_local_disabled
- get disabled ftrace_ops::disabled value for current cpu
Link: http://lkml.kernel.org/r/1329317514-8131-2-git-send-email-jolsa@redhat.com
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
If more than one __print_*() function is used in a tracepoint
(__print_flags(), __print_symbols(), etc), then the temp seq buffer will
not be zero on entry. Using the temp seq buffer's length to know if
data has been printed or not in the current function is incorrect and
may produce incorrect results.
Currently, no in-tree tracepoint causes this bug, but new ones may
be created.
Cc: Andrew Vagin <avagin@openvz.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
If __print_flags() is used after another __print_*() function, the
temp seq_file buffer will not be empty on entry, and the delimiter will
be printed even though there's just one field. We get something like:
|S
instead of just:
S
This is because the length of the temp seq buffer is used to determine
if the delimiter is printed or not. But this algorithm fails when
the seq buffer is not empty on entry, and the delimiter will be printed
because it thinks that a previous field was already printed.
Link: http://lkml.kernel.org/r/1329650167-480655-1-git-send-email-avagin@openvz.org
Signed-off-by: Andrew Vagin <avagin@openvz.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The __print_symbolic() function takes a sequence of key-value pairs for
pretty-printing a constant. The new kvm:kvm_exit print fmt uses the
expression:
__print_symbolic(..., { 0x040 + 1, "DB excp" }, ...)
Currently only atoms are supported and this print fmt fails to parse.
This patch adds support for expressions instead of just atoms so that
0x040 + 1 is parsed successfully. Also add arg_num_eval() support for
the '+' operator.
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1315148939-14313-1-git-send-email-stefanha@linux.vnet.ibm.com
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Includes smaller fixes and improvements plus the exclude_{host,guest} feature
test and fallback to handle older kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf_event_attr size needs to be initialized in all cases because it
captures the ABI version.
This patch moves the initialization of the field from the
perf_event_open() syscall stub to its proper location in the
event_attr_init().
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20120209151238.GA10272@quad
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
As the tracepoints in the cpuidle code are called when rcu_idle_exit() is in
effect, the _rcuidle() version must be used, otherwise the rcu_read_lock()s
that protect the tracepoint will not be honored.
Cc: Len Brown <len.brown@intel.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The power and cpuidle tracepoints are called within a rcu_idle_exit()
section, and must be denoted with the _rcuidle() version of the tracepoint.
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Added is a new static inline function that lets *any* tracepoint be used
inside a rcu_idle_exit() section. And this also solves the problem where
the same tracepoint may be used inside a rcu_idle_exit() section as well
as outside of one.
I added a new tracepoint function with a "_rcuidle" extension. All
tracepoints can be used with either the normal "trace_foobar()"
function, or the "trace_foobar_rcuidle()" function when inside a
rcu_idle_exit() section.
All tracepoints defined by TRACE_EVENT() or any of the derivatives
will have a "_rcuidle()" function also defined. When a tracepoint is
used within an rcu_idle_exit() section, the "_rcuidle()" version must
be used. This denotes that the tracepoint is within rcu_idle_exit()
and it allows the rcu read locks within the tracepoint to still
be valid, as this version takes us out of rcu_idle_exit().
Another nice aspect about this patch is that "static inline"s are not
compiled into text when not used. So only the tracepoints that actually
use the _rcuidle() version will have them defined in the actual text
that is booted.
Link: http://lkml.kernel.org/r/1328563113.2200.39.camel@gandalf.stny.rr.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Fix to decode grouped AVX with VEX pp bits which should be
handled as same as last-prefixes. This fixes below warnings
in posttest with CONFIG_CRYPTO_SHA1_SSSE3=y.
Warning: arch/x86/tools/test_get_len found difference at <sha1_transform_avx>:ffffffff810d5fc0
Warning: ffffffff810d6069: c5 f9 73 de 04 vpsrldq $0x4,%xmm6,%xmm0
Warning: objdump says 5 bytes, but insn_get_length() says 4
...
With this change, test_get_len can decode it correctly.
$ arch/x86/tools/test_get_len -v -y
ffffffff810d6069: c5 f9 73 de 04 vpsrldq $0x4,%xmm6,%xmm0
Succeed: decoded and checked 1 instructions
Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: yrl.pp-manager.tt@hitachi.com
Link: http://lkml.kernel.org/r/20120210053340.30429.73410.stgit@localhost.localdomain
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The current version of perf detects whether or not the perf.data file is
written in a different endianness using the attr_size field in the
header of the file. This field represents sizeof(struct perf_event_attr)
as known to perf record. If the sizes do not match, then perf tries the
byte-swapped version. If they match, then the tool assumes a different
endianness.
The issue with the approach is that it assumes the size of
perf_event_attr always has to match between perf record and perf report.
However, the kernel perf_event ABI is extensible. New fields can be
added to struct perf_event_attr. Consequently, it is not possible to use
attr_size to detect endianness.
This patch takes another approach by using the magic number written at
the beginning of the perf.data file to detect endianness. The magic
number is an eight-byte signature. It's primary purpose is to identify
(signature) a perf.data file. But it could also be used to encode the
endianness.
The patch introduces a new value for this signature. The key difference
is that the signature is written differently in the file depending on
the endianness. Thus, by comparing the signature from the file with the
tool's own signature it is possible to detect endianness. The new
signature is "PERFILE2".
Backward compatiblity with existing perf.data file is ensured.
Tested-by: David Ahern <dsahern@gmail.com>
Acked-by: David Ahern <dsahern@gmail.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Arun Sharma <asharma@fb.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Lin Ming <ming.m.lin@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roberto Agostino Vitillo <ravitillo@lbl.gov>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Vince Weaver <vweaver1@eecs.utk.edu>
Link: http://lkml.kernel.org/r/1328187288-24395-15-git-send-email-eranian@google.com
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Stephane Eranian reported that doing a scheduler latency
measurements with perf on AMD doesn't work out as expected due
to the fact that the sched_clock() granularity is too coarse,
i.e. done in jiffies due to the sched_clock_stable not set,
which, if set, would mean that we get to use the TSC as sample
source which would give us much higher precision.
However, there's no reason not to set sched_clock_stable on AMD
because all families from F10h and upwards do have an invariant
TSC and have the CPUID flag to prove (CPUID_8000_0007_EDX[8]).
Make it so, #1.
Signed-off-by: Borislav Petkov <bp@alien8.de>
Cc: Borislav Petkov <bp@amd64.org>
Cc: Venki Pallipadi <venki@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Andreas Herrmann <andreas.herrmann3@amd.com>
Link: http://lkml.kernel.org/r/20120206132546.GA30854@quad
[ Should any non-standard system break the TSC, we should
mark them so explicitly, in their platform init handler, or
in a DMI quirk. ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
