When we have a maximum latency reported for a task, we need a
convenient way to find the matching location to the raw traces
or to perf sched map that shows where the task has been
eventually scheduled in. This gives a pointer to retrieve the
events that occured during this max latency.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Reviewed-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1260391208-6808-1-git-send-regression-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Xiao Guangrong [Wed, 9 Dec 2009 09:51:30 +0000 (17:51 +0800)]
perf sched: Fix for getting task's execution time
In current code, task's execute time is got by reading
'/proc/<pid>/sched' file, it's wrong if the task is created
by pthread_create(), because every thread task has same pid.
This way also has two demerits:
1: 'perf sched replay' can't work if the kernel is not
compiled with the 'CONFIG_SCHED_DEBUG' option
2: perf tool should depend on proc file system
So, this patch uses PERF_COUNT_SW_TASK_CLOCK to get task's
execution time instead of reading /proc file.
Changelog v2 -> v3:
use PERF_COUNT_SW_TASK_CLOCK instead of rusage() as Ingo's
suggestion
tracing/kprobes: Fix field creation's bad error handling
When we define the common event fields in kprobe, we invert the error
handling and return immediately in case of success. Then we omit
to define specific kprobes fields (ip and nargs), and specific
kretprobes fields (func, ret_ip, nargs). And we only define them
when we fail to create common fields.
The most visible consequence is that we can't create filter for
k(ret)probes specific fields.
This patch re-invert the success/error handling to fix it.
Reported-by: Lai Jiangshan <laijs@cn.fujitsu.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Acked-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Li Zefan <lizf@cn.fujitsu.com>
LKML-Reference: <1260263815-5167-1-git-send-regression-fweisbec@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Xiao Guangrong [Wed, 9 Dec 2009 03:29:44 +0000 (11:29 +0800)]
perf_event: Allocate children's perf_event_ctxp at the right time
In current code, children task will allocate memory for
'child->perf_event_ctxp' if the parent is counted, we can
do it only if the parent allowed children inherit it.
hw-breakpoints: Modify breakpoints without unregistering them
Currently, when ptrace needs to modify a breakpoint, like disabling
it, changing its address, type or len, it calls
modify_user_hw_breakpoint(). This latter will perform the heavy and
racy task of unregistering the old breakpoint and registering a new
one.
This is racy as someone else might steal the reserved breakpoint
slot under us, which is undesired as the breakpoint is only
supposed to be modified, sometimes in the middle of a debugging
workflow. We don't want our slot to be stolen in the middle.
So instead of unregistering/registering the breakpoint, just
disable it while we modify its breakpoint fields and re-enable it
after if necessary.
Support perf probe --del <event> option. Currently,
perf probe can have only one event for each --del option.
If you'd like to delete several probe events, you need
to specify --del for each events.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Jim Keniston <jkenisto@us.ibm.com> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jason Baron <jbaron@redhat.com> Cc: K.Prasad <prasad@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net>
LKML-Reference: <20091208220323.10142.62079.stgit@dhcp-100-2-132.bos.redhat.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Remove event suffix number _0 if it is the first.
The first event has no suffix, and from the second,
each event has suffix number counted from _1. This
reduces typing cost :-).
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Jim Keniston <jkenisto@us.ibm.com> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jason Baron <jbaron@redhat.com> Cc: K.Prasad <prasad@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net>
LKML-Reference: <20091208220301.10142.50031.stgit@dhcp-100-2-132.bos.redhat.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf probe: Fix add-probe command syntax without --add option
Fix add-probe command syntax without --add option.
perf-probe supports add-probe command without --add
option. But it treats each argument as an event definition.
e.g.
perf probe func arg1 arg2
is interpreted as
perf probe --add func --add arg1 --add arg2
But it may be useless in many cases.
This patch fixes this syntax to fold those arguments into
one event definition if there is no --add option. With this
change, above command is interpreted as below;
perf probe --add "func arg1 arg2"
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Jim Keniston <jkenisto@us.ibm.com> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jason Baron <jbaron@redhat.com> Cc: K.Prasad <prasad@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net>
LKML-Reference: <20091208220254.10142.73767.stgit@dhcp-100-2-132.bos.redhat.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Change event list format for user readability. perf probe --list
shows event list in "[GROUP:EVENT] EVENT-DEFINITION" format, but
this format is different from the output of perf-list, and
EVENT-DEFINITION is a bit blunt. This patch changes the format to
more user friendly one.
Before:
[probe:schedule_0] schedule+10 prev cpu
After:
probe:schedule_0 (on schedule+10 with prev cpu)
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Jim Keniston <jkenisto@us.ibm.com> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jason Baron <jbaron@redhat.com> Cc: K.Prasad <prasad@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net>
LKML-Reference: <20091208220240.10142.42916.stgit@dhcp-100-2-132.bos.redhat.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf probe: Fix event namelist to duplicate string
Fix event namelist to duplicate string. Without duplicating, adding
multiple probes causes stack overwrite bug, because it reuses a
buffer on stack while the buffer is already added in the namelist.
String duplication solves this bug because only contents of the
buffer is copied to the namelist.
Uses of strcat are almost always signs that someone is too lazy
to think about the code a bit more carefully. One always has to
know about the lengths of the strings involved to avoid buffer
overflows.
This is one case where the size of the object code for me is
reduced by 38 bytes. The code should also be faster, especially
if flags is non-NULL.
Whatever the context nature of a breakpoint, we always perform the
following constraint checks before allocating it a slot:
- Check the number of pinned breakpoint bound the concerned cpus
- Check the max number of task-bound breakpoints that are belonging
to a task.
- Add both and see if we have a reamining slot for the new breakpoint
This is the right thing to do when we are about to register a cpu-only
bound breakpoint. But not if we are dealing with a task bound
breakpoint. What we want in this case is:
- Check the number of pinned breakpoint bound the concerned cpus
- Check the number of breakpoints that already belong to the task
in which the breakpoint to register is bound to.
- Add both
This fixes a regression that makes the "firefox -g" command fail to
register breakpoints once we deal with a secondary thread.
Reported-by: Walt <w41ter@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Prasad <prasad@linux.vnet.ibm.com>
hw-breakpoints: Zeroe the breakpoint attrs on initialization
The perf attrs used to set up breakpoint parameters are often allocated
in the stack and not zeroed out before calling hw_breakpoint_init().
Handle it from this helper to avoid random attributes set by the stack.
OGAWA Hirofumi [Sun, 6 Dec 2009 11:08:24 +0000 (20:08 +0900)]
perf: Make common SAMPLE_EVENT parser
Currently, sample event data is parsed for each commands, and it
is assuming that the data is not including other data. (E.g.
timechart, trace, etc. can't parse the event if it has
PERF_SAMPLE_CALLCHAIN)
So, even if we record the superset data for multiple commands at
a time, commands can't parse. etc.
To fix it, this makes common sample event parser, and use it to
parse sample event correctly. (PERF_SAMPLE_READ is unsupported
for now though, it seems to be not using.)
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <87hbs48imv.fsf@devron.myhome.or.jp> Signed-off-by: Ingo Molnar <mingo@elte.hu>
OGAWA Hirofumi [Sun, 6 Dec 2009 11:07:29 +0000 (20:07 +0900)]
perf timechart: Fix header handling
Update "struct trace_entry" to match with current one. And
remove "size" field from it.
If it has "size", it become cause of alignment mismatch of
structure with kernel.
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp> Acked-by: Arjan van de Ven <arjan@infradead.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <87ljhg8ioe.fsf@devron.myhome.or.jp> Signed-off-by: Ingo Molnar <mingo@elte.hu>
When we enter in irq, two things can happen to preserve the link
to the previous frame pointer:
- If we were in an irq already, we don't switch to the irq stack
as we are inside. We just need to save the previous frame
pointer and to link the new one to the previous.
- Otherwise we need another level of indirection. We enter the irq with
the previous stack. We save the previous bp inside and make bp
pointing to its saved address. Then we switch to the irq stack and
push bp another time but to the new stack. This makes two levels to
dereference instead of one.
In the second case, the current stacktrace code omits the second level
and loses the frame pointer accuracy. The stack that follows will then
be considered as unreliable.
Handling that makes the perf callchain happier.
Before:
x86: Fixup wrong debug exception frame link in stacktraces
While dumping a stacktrace, the end of the exception stack won't link
the frame pointer to the previous stack.
The interrupted stack will then be considered as unreliable and ignored
by perf, as the frame pointer is unreliable itself.
This happens because we overwrite the frame pointer that links to the
interrupted frame with the address of the exception stack. This is
done in order to reserve space inside.
But rbp has been chosen here only because it is not a scratch register,
so that the address of the exception stack remains in rbp after calling
do_debug(), we can then release the exception stack space without the
need to retrieve its address again.
But we can pick another non-scratch register to do that, so that we
preserve the link to the interrupted stack frame in the stacktraces.
Just randomly choose r12. Every registers are saved just before and
restored just after calling do_debug(). And r12 is not used in the
middle, which makes it a perfect candidate.
Example: perf record -g -a -c 1 -f -e mem:$(tasklist_lock_addr):rw
We are infected with all the debug stack. Like the nmi stack, the debug
stack is undesired as it is part of the profiling path, not helpful for
the user.
Ignore it.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: "K. Prasad" <prasad@linux.vnet.ibm.com>
hw-breakpoints: Use overflow handler instead of the event callback
struct perf_event::event callback was called when a breakpoint
triggers. But this is a rather opaque callback, pretty
tied-only to the breakpoint API and not really integrated into perf
as it triggers even when we don't overflow.
We prefer to use overflow_handler() as it fits into the perf events
rules, being called only when we overflow.
Reported-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: "K. Prasad" <prasad@linux.vnet.ibm.com>
hw-breakpoints: Drop callback and task parameters from modify helper
Drop the callback and task parameters from modify_user_hw_breakpoint().
For now we have no user that need to modify a breakpoint to the point
of changing its handler or its task context.
Jason Baron [Tue, 1 Dec 2009 17:18:49 +0000 (12:18 -0500)]
tracing: Add DEFINE_EVENT(), DEFINE_SINGLE_EVENT() support to docbook
The introduction of the new 'DECLARE_EVENT_CLASS()' obviates the
need for the 'TRACE_EVENT()' macro in some cases. Thus, docbook
style comments that used to live with 'TRACE_EVENT()' are now
moved to 'DEFINE_EVENT()'. Thus, we need to make the docbook
system understand the new 'DEFINE_EVENT()' macro. In addition
I've tried to futureproof the patch, by also adding support for
'DEFINE_SINGLE_EVENT()', since there has been discussion about
renaming: TRACE_EVENT() -> DEFINE_SINGLE_EVENT().
Without this patch the tracepoint docbook fails to build.
I've verified that this patch correctly builds the tracepoint
docbook which currently covers signals, and irqs.
Changes in v2:
- properly indent perl 'if' statements
Signed-off-by: Jason Baron <jbaron@redhat.com> Acked-by: Steven Rostedt <rostedt@goodmis.org> Acked-by: Randy Dunlap <randy.dunlap@oracle.com> Cc: William Cohen <wcohen@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> Cc: Masami Hiramatsu <mhiramat@redhat.com>
LKML-Reference: <200912011718.nB1HIn7t011371@int-mx04.intmail.prod.int.phx2.redhat.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf: Don't free perf_mmap_data until work has been done
In the CONFIG_PERF_USE_VMALLOC case, perf_mmap_data_free() only
schedules the cleanup of the perf_mmap_data struct. In that
case we have to wait until the work has been done before we free
data.
Signed-off-by: Kristian Høgsberg <krh@bitplanet.net> Cc: David S. Miller <davem@davemloft.net> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: <stable@kernel.org>
LKML-Reference: <1259697901-1747-1-git-send-email-krh@bitplanet.net> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Simplify event naming as <symbol>_<seqnum>. Each event name is
globally unique (group name is not checked). So, if there is
schedule_0, next probe event on schedule() will be schedule_1.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Jim Keniston <jkenisto@us.ibm.com> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jason Baron <jbaron@redhat.com> Cc: K.Prasad <prasad@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20091201002024.10235.2353.stgit@harusame> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf probe: Change a debugging message from pr_info to pr_debug
Change annoying debug-info using notice from pr_info() to
pr_debug(), since the message always printed when user adds a
probe point which requires debug-info.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Jim Keniston <jkenisto@us.ibm.com> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jason Baron <jbaron@redhat.com> Cc: K.Prasad <prasad@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20091201001927.10235.63645.stgit@harusame> Signed-off-by: Ingo Molnar <mingo@elte.hu>
trace_kprobes: Fix a memory leak bug and check kstrdup() return value
Fix a memory leak case in create_trace_probe(). When an argument
is too long (> MAX_ARGSTR_LEN), it just jumps to error path. In
that case tp->args[i].name is not released.
This also fixes a bug to check kstrdup()'s return value.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: systemtap <systemtap@sources.redhat.com> Cc: DLE <dle-develop@lists.sourceforge.net> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Jim Keniston <jkenisto@us.ibm.com> Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Frank Ch. Eigler <fche@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jason Baron <jbaron@redhat.com> Cc: K.Prasad <prasad@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <20091201001919.10235.56455.stgit@harusame> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Convert builtin-timechart.c to mmap_dispatch_perf_file() +
perf_file_handler.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com> Acked-by: Arjan van de Ven <arjan@infradead.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Peter Zijlstra <peterz@infradead.org>
LKML-Reference: <4B14B21C.2040406@cn.fujitsu.com>
[ v2: cleaned up the printout, fixed a whitespace detail ] Signed-off-by: Ingo Molnar <mingo@elte.hu>
Li Zefan [Tue, 1 Dec 2009 06:04:49 +0000 (14:04 +0800)]
perf annotate: Fix perf data parsing
perf-annotate doesn't parse perf.data correctly in that it
doesn't read perf header. Fix this by using
mmap_dispatch_perf_file().
Before:
TOTAL events: 17565
MMAP events: 3221
LOST events: 10
COMM events: 235
EXIT events: 2
THROTTLE events: 1
UNTHROTTLE events: 2
FORK events: 10
READ events: 1
SAMPLE events: 14083
After:
TOTAL events: 17290
MMAP events: 3203
LOST events: 0
COMM events: 234
EXIT events: 1
THROTTLE events: 0
UNTHROTTLE events: 0
FORK events: 0
READ events: 0
SAMPLE events: 13852
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Arjan van de Ven <arjan@infradead.org>
LKML-Reference: <4B14B201.9030708@cn.fujitsu.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
The common_* functions (e.g. common_pc(), etc) are exported as
common_* but named get_common_*, resulting in unresolved
subroutine errors when executing scripts.
Tom Zanussi [Mon, 30 Nov 2009 07:18:47 +0000 (01:18 -0600)]
perf trace/scripting: Ignore shadowed variable warning for perf-trace-perl.c
The debugging versions of the ENTER and LEAVE internal perl
macros, used when embedding perl, define a local block with a
my_perl perl variable that shadows a global variable of the same
name, which is also the name expected by the embedding API for
the embedded interpreter.
Since we don't have control over the code generated in this case
(it's an externality) and can't get rid of the warning, ignore it.
Tom Zanussi [Wed, 25 Nov 2009 07:15:52 +0000 (01:15 -0600)]
perf trace: Add a scripts/perl/bin for perf trace shell scripts
To capture the relevant events for a given Perl script and to
avoid having to continually remember and type in long
command-lines, add a scripts/perl/bin directory containing two
simple shell scripts for each Perl script, one for recording and
one for processing/display. For example, to record perf data for
the rw-by-pid.pl script, run scripts/perl/bin/rw-by-pid-record
and to actually run the script and display the output run
scripts/perl/bin/rw-by-pid-report.
Tom Zanussi [Wed, 25 Nov 2009 07:15:50 +0000 (01:15 -0600)]
perf trace: Add interface to access perf data from Perl handlers
The Perl scripting support for perf trace allows most of a trace
event's data to be accessed directly as handler arguments, but
not all of it e.g. the less common fields aren't passed in. To
give scripts access to the other fields and/or any other data or
metadata in the main perf executable that might be useful, a way
to access the C data in perf from Perl is needed; this patch
uses the Perl XS facility to do it for the common_xxx event
fields not passed to handler functions.
Context.pm exports three functions to Perl scripts that access
fields for the current event by calling back into perf:
common_pc(), common_flags() and common_lock_depth(). Support
for common_flags() field values was added to Core.pm and a
script used to sanity check these and other basic scripting
features, check-perf-trace.pl, was also added.
Tom Zanussi [Wed, 25 Nov 2009 07:15:49 +0000 (01:15 -0600)]
perf trace: Add perf trace scripting support modules for Perl
Add Perf-Trace-Util Perl module and some scripts that use it.
Core.pm contains Perl code to define and access flag and
symbolic fields. Util.pm contains general-purpose utility
functions.
Also adds some makefile bits to install them in
libexec/perf-core/scripts/perl (or wherever perfexec_instdir
points).
Tom Zanussi [Wed, 25 Nov 2009 07:15:48 +0000 (01:15 -0600)]
perf trace: Add Perl scripting support
Implement trace_scripting_ops to make Perl a supported perf
trace scripting language.
Additionally adds code that allows Perl trace scripts to access
the 'flag' and 'symbolic' (__print_flags(), __print_symbolic())
field information parsed from the trace format files.
Also adds the Perl implementation of the generate_script()
trace_scripting_op, which creates a ready-to-run perf trace Perl
script based on existing trace data. Scripts generated by this
implementation print out all the fields for each event mentioned
in perf.data (and will detect and generate the proper scripting
code for 'flag' and 'symbolic' fields), and will additionally
generate handlers for the special 'trace_unhandled',
'trace_begin' and 'trace_end' handlers. Script authors can
simply remove the printing code to implement their own custom
event handling.
Tom Zanussi [Wed, 25 Nov 2009 07:15:47 +0000 (01:15 -0600)]
perf trace: Add flag/symbolic format_flags
It's useful to know whether a field is a flag or symbolic field
for e.g. when generating scripts - it allows us to translate
those fields specially rather than literally as plain numeric
values.
Tom Zanussi [Wed, 25 Nov 2009 07:15:46 +0000 (01:15 -0600)]
perf trace: Add scripting ops
Adds an interface, scripting_ops, that when implemented for a
particular scripting language enables built-in support for trace
stream processing using that language.
The interface is designed to enable full-fledged language
interpreters to be embedded inside the perf executable and
thereby make the full capabilities of the supported languages
available for trace processing.
See below for details on the interface.
This patch also adds a couple command-line options to 'perf
trace':
The -s option option is used to specify the script to be run.
Script names that can be used with -s take the form:
[language spec:]scriptname[.ext]
Scripting languages register a set of 'language specs' that can
be used to specify scripts for the registered languages. The
specs can be used either as prefixes or extensions.
If [language spec:] is used, the script is taken as a script of
the matching language regardless of any extension it might have.
If [language spec:] is not used, [.ext] is used to look up the
language it corresponds to. Language specs are case
insensitive.
e.g. Perl scripts can be specified in the following ways:
The -g [language spec] option gives users an easy starting point
for writing scripts in the specified language. Scripting
support for a particular language can implement a
generate_script() scripting op that outputs an empty (or
near-empty) set of handlers for all the events contained in a
given perf.data trace file - this option gives users a direct
way to access that.
Adding support for a scripting language
---------------------------------------
The main thing that needs to be done do add support for a new
language is to implement the scripting_ops interface:
start_script() is called before any events are processed, and is
meant to give the scripting language support an opportunity to
set things up to receive events e.g. create and initialize an
instance of a language interpreter.
stop_script() is called after all events are processed, and is
meant to give the scripting language support an opportunity to
clean up e.g. destroy the interpreter instance, etc.
process_event() is called once for each event and takes as its
main parameter a pointer to the binary trace event record to be
processed. The implementation is responsible for picking out the
binary fields from the event record and sending them to the
script handler function associated with that event e.g. a
function derived from the event name it's meant to handle e.g.
'sched::sched_switch()'. The 'format' information for trace
events can be used to parse the binary data and map it into a
form usable by a given scripting language; see the Perl
implemention in subsequent patches for one possible way to
leverage the existing trace format parsing code in perf and map
that info into specific scripting language types.
generate_script() should generate a ready-to-run script for the
current set of events in the trace, preferably with bodies that
print out every field for each event. Again, look at the Perl
implementation for clues as to how that can be done. This is an
optional, but very useful op.
Support for a given language should also add a language-specific
setup function and call it from setup_scripting(). The
language-specific setup function associates the the scripting
ops for that language with one or more 'language specifiers'
(see below) using script_spec_register(). When a script name is
specified on the command line, the scripting ops associated with
the specified language are used to instantiate and use the
appropriate interpreter to process the trace stream.
In general, it should be relatively easy to add support for a
new language, especially if the language implementation supports
an interface allowing an interpreter to be 'embedded' inside
another program (in this case the containing program will be
'perf trace'). If so, it should be relatively straightforward to
translate trace events into invocations of user-defined script
functions where e.g. the function name corresponds to the event
type and the function parameters correspond to the event fields.
The event and field type information exported by the event
tracing infrastructure (via the event 'format' files) should be
enough to parse and send any piece of trace data to the user
script. The easiest way to see how this can be done would be to
look at the Perl implementation contained in
perf/util/trace-event-perl.c/.h.
There are a couple of other things that aren't covered by the
scripting_ops or setup interface and are technically optional,
but should be implemented if possible. One of these is support
for 'flag' and 'symbolic' fields e.g. being able to use more
human-readable values such as 'GFP_KERNEL' or
HI/BLOCK_IOPOLL/TASKLET in place of raw flag values. See the
Perl implementation to see how this can be done. The other thing
is support for 'calling back' into the perf executable to access
e.g. uncommon fields not passed by default into handler
functions, or any metadata the implementation might want to make
available to users via the language interface. Again, see the
Perl implementation for examples.
perf tools: Consolidate symbol resolving across all tools
Now we have a very high level routine for simple tools to
process IP sample events:
int event__preprocess_sample(const event_t *self,
struct addr_location *al,
symbol_filter_t filter)
It receives the event itself and will insert new threads in the
global threads list and resolve the map and symbol, filling all
this info into the new addr_location struct, so that tools like
annotate and report can further process the event by creating
hist_entries in their specific way (with or without callgraphs,
etc).
This one will, given a thread (userspace or the kernel kthread
one), will find the given type (MAP__FUNCTION now, MAP__VARIABLE
too in the near future) at the given cpumode, taking vdsos into
account (userspace hit, but kernel symbol) and will fill all
these details in the addr_location given.
Tools that need a more compact API for plain function
resolution, like 'kmem', can use this other one:
struct symbol *thread__find_function(struct thread *self, u64 addr,
symbol_filter_t filter)
So, to resolve a kernel symbol, that is all the 'kmem' tool
needs, its just a matter of calling:
sym = thread__find_function(kthread, addr, NULL);
The 'filter' parameter is needed because we do lazy
parsing/loading of ELF symtabs or /proc/kallsyms.
With this we remove more code duplication all around, which is
always good, huh? :-)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: John Kacur <jkacur@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1259346563-12568-12-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
While implementing event__preprocess_sample, that will do all of
the symbol lookup in one convenient function, I noticed that
util/process_event.[ch] were not being used at all, then started
looking if there were other functions that could be shared
and...
All those functions really don't need to receive offset + head,
the only thing they did was common to all of them, so do it at
one place instead.
Stats about number of each type of event processed now is done
in a central place.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: John Kacur <jkacur@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1259346563-12568-11-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf symbols: Support multiple symtabs in struct thread
Making the routines that were so far specific to the kernel maps
useful for all threads.
This is done by making the kernel maps be contained in a kernel
"thread".
This gets the kernel specific routines closer to the userspace
counterparts, which will help in reducing the boilerplate for
resolving a symbol, as will be demonstrated in the next patches.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1259346563-12568-9-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf symbols: Make the kallsyms loading routines part of the dso class
So that the kallsyms loading routines are the direct counterpart
of the vmlinux loading ones, i.e. dso__load_kallsyms is the
counterpart of dso__load_vmlinux.
In the process make them also use the symbols rb tree indexed by
map->type, paving the way for supporting other types of symtabs,
such as the next one to be supported: variables.
This also allowed removal of yet another global variable:
kernel_map__functions.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1259346563-12568-7-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
perf symbols: Better support for multiple symbol tables per dso
By using an array of rb_roots in struct dso we can, from a
struct map instance to get the right symbol rb_tree more easily.
This way we can have just one symbol lookup method for struct
map instances, map__find_symbol, instead of one per symtab type
(functions, variables).
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1259346563-12568-6-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
projects / kernel.git / log