2009-06-02 22:59:57 +02:00
|
|
|
/*
|
2009-06-02 23:37:05 +02:00
|
|
|
* builtin-record.c
|
|
|
|
*
|
|
|
|
* Builtin record command: Record the profile of a workload
|
|
|
|
* (or a CPU, or a PID) into the perf.data output file - for
|
|
|
|
* later analysis via perf report.
|
2009-06-02 22:59:57 +02:00
|
|
|
*/
|
2009-05-27 09:10:38 +02:00
|
|
|
#include "builtin.h"
|
2009-06-02 23:37:05 +02:00
|
|
|
|
|
|
|
#include "perf.h"
|
|
|
|
|
2009-05-01 18:29:57 +02:00
|
|
|
#include "util/util.h"
|
2009-05-26 09:17:18 +02:00
|
|
|
#include "util/parse-options.h"
|
2009-05-26 11:10:09 +02:00
|
|
|
#include "util/parse-events.h"
|
2009-06-01 22:50:19 +02:00
|
|
|
#include "util/string.h"
|
2009-05-01 18:29:57 +02:00
|
|
|
|
2009-06-25 17:05:54 +02:00
|
|
|
#include "util/header.h"
|
2009-08-12 11:07:25 +02:00
|
|
|
#include "util/event.h"
|
2009-08-16 22:05:48 +02:00
|
|
|
#include "util/debug.h"
|
2009-12-12 00:24:02 +01:00
|
|
|
#include "util/session.h"
|
perf symbols: Use the buildids if present
With this change 'perf record' will intercept PERF_RECORD_MMAP
calls, creating a linked list of DSOs, then when the session
finishes, it will traverse this list and read the buildids,
stashing them at the end of the file and will set up a new
feature bit in the header bitmask.
'perf report' will then notice this feature and populate the
'dsos' list and set the build ids.
When reading the symtabs it will refuse to load from a file that
doesn't have the same build id. This improves the
reliability of the profiler output, as symbols and profiling
data is more guaranteed to match.
Example:
[root@doppio ~]# perf report | head
/home/acme/bin/perf with build id b1ea544ac3746e7538972548a09aadecc5753868 not found, continuing without symbols
# Samples: 2621434559
#
# Overhead Command Shared Object Symbol
# ........ ............... ............................. ......
#
7.91% init [kernel] [k] read_hpet
7.64% init [kernel] [k] mwait_idle_with_hints
7.60% swapper [kernel] [k] read_hpet
7.60% swapper [kernel] [k] mwait_idle_with_hints
3.65% init [kernel] [k] 0xffffffffa02339d9
[root@doppio ~]#
In this case the 'perf' binary was an older one, vanished,
so its symbols probably wouldn't match or would cause subtly
different (and misleading) output.
Next patches will support the kernel as well, reading the build
id notes for it and the modules from /sys.
Another patch should also introduce a new plumbing command:
'perf list-buildids'
that will then be used in porcelain that is distro specific to
fetch -debuginfo packages where such buildids are present. This
will in turn allow for one to run 'perf record' in one machine
and 'perf report' in another.
Future work on having the buildid sent directly from the kernel
in the PERF_RECORD_MMAP event is needed to close races, as the
DSO can be changed during a 'perf record' session, but this
patch at least helps with non-corner cases and current/older
kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.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: Jim Keniston <jkenisto@us.ibm.com>
Cc: K. Prasad <prasad@linux.vnet.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1257367843-26224-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-11-04 21:50:43 +01:00
|
|
|
#include "util/symbol.h"
|
2009-06-25 17:05:54 +02:00
|
|
|
|
2009-06-02 15:52:24 +02:00
|
|
|
#include <unistd.h>
|
2009-04-08 15:01:31 +02:00
|
|
|
#include <sched.h>
|
|
|
|
|
|
|
|
static int fd[MAX_NR_CPUS][MAX_COUNTERS];
|
2009-06-06 09:58:57 +02:00
|
|
|
|
2009-10-12 07:56:03 +02:00
|
|
|
static long default_interval = 0;
|
2009-06-06 09:58:57 +02:00
|
|
|
|
2009-10-06 15:14:21 +02:00
|
|
|
static int nr_cpus = 0;
|
2009-04-08 15:01:31 +02:00
|
|
|
static unsigned int page_size;
|
2009-10-06 15:14:21 +02:00
|
|
|
static unsigned int mmap_pages = 128;
|
|
|
|
static int freq = 1000;
|
2009-04-08 15:01:31 +02:00
|
|
|
static int output;
|
2009-05-27 09:33:18 +02:00
|
|
|
static const char *output_name = "perf.data";
|
2009-10-06 15:14:21 +02:00
|
|
|
static int group = 0;
|
|
|
|
static unsigned int realtime_prio = 0;
|
|
|
|
static int raw_samples = 0;
|
|
|
|
static int system_wide = 0;
|
|
|
|
static int profile_cpu = -1;
|
|
|
|
static pid_t target_pid = -1;
|
|
|
|
static pid_t child_pid = -1;
|
|
|
|
static int inherit = 1;
|
|
|
|
static int force = 0;
|
|
|
|
static int append_file = 0;
|
|
|
|
static int call_graph = 0;
|
|
|
|
static int inherit_stat = 0;
|
|
|
|
static int no_samples = 0;
|
|
|
|
static int sample_address = 0;
|
|
|
|
static int multiplex = 0;
|
|
|
|
static int multiplex_fd = -1;
|
|
|
|
|
|
|
|
static long samples = 0;
|
2009-06-06 09:58:57 +02:00
|
|
|
static struct timeval last_read;
|
|
|
|
static struct timeval this_read;
|
|
|
|
|
2009-10-06 15:14:21 +02:00
|
|
|
static u64 bytes_written = 0;
|
2009-06-06 09:58:57 +02:00
|
|
|
|
|
|
|
static struct pollfd event_array[MAX_NR_CPUS * MAX_COUNTERS];
|
|
|
|
|
2009-10-06 15:14:21 +02:00
|
|
|
static int nr_poll = 0;
|
|
|
|
static int nr_cpu = 0;
|
2009-06-06 09:58:57 +02:00
|
|
|
|
2009-10-06 15:14:21 +02:00
|
|
|
static int file_new = 1;
|
2009-06-25 17:05:54 +02:00
|
|
|
|
2009-12-12 00:24:02 +01:00
|
|
|
static struct perf_session *session;
|
2009-06-18 23:22:55 +02:00
|
|
|
|
2009-04-08 15:01:31 +02:00
|
|
|
struct mmap_data {
|
2009-06-06 09:58:57 +02:00
|
|
|
int counter;
|
|
|
|
void *base;
|
|
|
|
unsigned int mask;
|
|
|
|
unsigned int prev;
|
2009-04-08 15:01:31 +02:00
|
|
|
};
|
|
|
|
|
2009-06-06 09:58:57 +02:00
|
|
|
static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS];
|
|
|
|
|
2009-06-18 11:40:28 +02:00
|
|
|
static unsigned long mmap_read_head(struct mmap_data *md)
|
2009-04-08 15:01:31 +02:00
|
|
|
{
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
struct perf_event_mmap_page *pc = md->base;
|
2009-06-18 11:40:28 +02:00
|
|
|
long head;
|
2009-04-08 15:01:31 +02:00
|
|
|
|
|
|
|
head = pc->data_head;
|
|
|
|
rmb();
|
|
|
|
|
|
|
|
return head;
|
|
|
|
}
|
|
|
|
|
2009-06-18 11:40:28 +02:00
|
|
|
static void mmap_write_tail(struct mmap_data *md, unsigned long tail)
|
|
|
|
{
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
struct perf_event_mmap_page *pc = md->base;
|
2009-06-18 11:40:28 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* ensure all reads are done before we write the tail out.
|
|
|
|
*/
|
|
|
|
/* mb(); */
|
|
|
|
pc->data_tail = tail;
|
|
|
|
}
|
|
|
|
|
2009-06-18 23:22:55 +02:00
|
|
|
static void write_output(void *buf, size_t size)
|
|
|
|
{
|
|
|
|
while (size) {
|
|
|
|
int ret = write(output, buf, size);
|
|
|
|
|
|
|
|
if (ret < 0)
|
|
|
|
die("failed to write");
|
|
|
|
|
|
|
|
size -= ret;
|
|
|
|
buf += ret;
|
|
|
|
|
|
|
|
bytes_written += ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
perf symbols: Use the buildids if present
With this change 'perf record' will intercept PERF_RECORD_MMAP
calls, creating a linked list of DSOs, then when the session
finishes, it will traverse this list and read the buildids,
stashing them at the end of the file and will set up a new
feature bit in the header bitmask.
'perf report' will then notice this feature and populate the
'dsos' list and set the build ids.
When reading the symtabs it will refuse to load from a file that
doesn't have the same build id. This improves the
reliability of the profiler output, as symbols and profiling
data is more guaranteed to match.
Example:
[root@doppio ~]# perf report | head
/home/acme/bin/perf with build id b1ea544ac3746e7538972548a09aadecc5753868 not found, continuing without symbols
# Samples: 2621434559
#
# Overhead Command Shared Object Symbol
# ........ ............... ............................. ......
#
7.91% init [kernel] [k] read_hpet
7.64% init [kernel] [k] mwait_idle_with_hints
7.60% swapper [kernel] [k] read_hpet
7.60% swapper [kernel] [k] mwait_idle_with_hints
3.65% init [kernel] [k] 0xffffffffa02339d9
[root@doppio ~]#
In this case the 'perf' binary was an older one, vanished,
so its symbols probably wouldn't match or would cause subtly
different (and misleading) output.
Next patches will support the kernel as well, reading the build
id notes for it and the modules from /sys.
Another patch should also introduce a new plumbing command:
'perf list-buildids'
that will then be used in porcelain that is distro specific to
fetch -debuginfo packages where such buildids are present. This
will in turn allow for one to run 'perf record' in one machine
and 'perf report' in another.
Future work on having the buildid sent directly from the kernel
in the PERF_RECORD_MMAP event is needed to close races, as the
DSO can be changed during a 'perf record' session, but this
patch at least helps with non-corner cases and current/older
kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.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: Jim Keniston <jkenisto@us.ibm.com>
Cc: K. Prasad <prasad@linux.vnet.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1257367843-26224-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-11-04 21:50:43 +01:00
|
|
|
static void write_event(event_t *buf, size_t size)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Add it to the list of DSOs, so that when we finish this
|
|
|
|
* record session we can pick the available build-ids.
|
|
|
|
*/
|
|
|
|
if (buf->header.type == PERF_RECORD_MMAP)
|
|
|
|
dsos__findnew(buf->mmap.filename);
|
|
|
|
|
|
|
|
write_output(buf, size);
|
|
|
|
}
|
|
|
|
|
2009-10-26 22:23:18 +01:00
|
|
|
static int process_synthesized_event(event_t *event)
|
|
|
|
{
|
perf symbols: Use the buildids if present
With this change 'perf record' will intercept PERF_RECORD_MMAP
calls, creating a linked list of DSOs, then when the session
finishes, it will traverse this list and read the buildids,
stashing them at the end of the file and will set up a new
feature bit in the header bitmask.
'perf report' will then notice this feature and populate the
'dsos' list and set the build ids.
When reading the symtabs it will refuse to load from a file that
doesn't have the same build id. This improves the
reliability of the profiler output, as symbols and profiling
data is more guaranteed to match.
Example:
[root@doppio ~]# perf report | head
/home/acme/bin/perf with build id b1ea544ac3746e7538972548a09aadecc5753868 not found, continuing without symbols
# Samples: 2621434559
#
# Overhead Command Shared Object Symbol
# ........ ............... ............................. ......
#
7.91% init [kernel] [k] read_hpet
7.64% init [kernel] [k] mwait_idle_with_hints
7.60% swapper [kernel] [k] read_hpet
7.60% swapper [kernel] [k] mwait_idle_with_hints
3.65% init [kernel] [k] 0xffffffffa02339d9
[root@doppio ~]#
In this case the 'perf' binary was an older one, vanished,
so its symbols probably wouldn't match or would cause subtly
different (and misleading) output.
Next patches will support the kernel as well, reading the build
id notes for it and the modules from /sys.
Another patch should also introduce a new plumbing command:
'perf list-buildids'
that will then be used in porcelain that is distro specific to
fetch -debuginfo packages where such buildids are present. This
will in turn allow for one to run 'perf record' in one machine
and 'perf report' in another.
Future work on having the buildid sent directly from the kernel
in the PERF_RECORD_MMAP event is needed to close races, as the
DSO can be changed during a 'perf record' session, but this
patch at least helps with non-corner cases and current/older
kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.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: Jim Keniston <jkenisto@us.ibm.com>
Cc: K. Prasad <prasad@linux.vnet.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1257367843-26224-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-11-04 21:50:43 +01:00
|
|
|
write_event(event, event->header.size);
|
2009-10-26 22:23:18 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-04-08 15:01:31 +02:00
|
|
|
static void mmap_read(struct mmap_data *md)
|
|
|
|
{
|
|
|
|
unsigned int head = mmap_read_head(md);
|
|
|
|
unsigned int old = md->prev;
|
|
|
|
unsigned char *data = md->base + page_size;
|
|
|
|
unsigned long size;
|
|
|
|
void *buf;
|
|
|
|
int diff;
|
|
|
|
|
|
|
|
gettimeofday(&this_read, NULL);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we're further behind than half the buffer, there's a chance
|
2009-06-05 14:29:10 +02:00
|
|
|
* the writer will bite our tail and mess up the samples under us.
|
2009-04-08 15:01:31 +02:00
|
|
|
*
|
|
|
|
* If we somehow ended up ahead of the head, we got messed up.
|
|
|
|
*
|
|
|
|
* In either case, truncate and restart at head.
|
|
|
|
*/
|
|
|
|
diff = head - old;
|
2009-06-18 11:40:28 +02:00
|
|
|
if (diff < 0) {
|
2009-04-08 15:01:31 +02:00
|
|
|
struct timeval iv;
|
|
|
|
unsigned long msecs;
|
|
|
|
|
|
|
|
timersub(&this_read, &last_read, &iv);
|
|
|
|
msecs = iv.tv_sec*1000 + iv.tv_usec/1000;
|
|
|
|
|
|
|
|
fprintf(stderr, "WARNING: failed to keep up with mmap data."
|
|
|
|
" Last read %lu msecs ago.\n", msecs);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* head points to a known good entry, start there.
|
|
|
|
*/
|
|
|
|
old = head;
|
|
|
|
}
|
|
|
|
|
|
|
|
last_read = this_read;
|
|
|
|
|
|
|
|
if (old != head)
|
2009-06-05 14:29:10 +02:00
|
|
|
samples++;
|
2009-04-08 15:01:31 +02:00
|
|
|
|
|
|
|
size = head - old;
|
|
|
|
|
|
|
|
if ((old & md->mask) + size != (head & md->mask)) {
|
|
|
|
buf = &data[old & md->mask];
|
|
|
|
size = md->mask + 1 - (old & md->mask);
|
|
|
|
old += size;
|
2009-06-03 19:27:19 +02:00
|
|
|
|
perf symbols: Use the buildids if present
With this change 'perf record' will intercept PERF_RECORD_MMAP
calls, creating a linked list of DSOs, then when the session
finishes, it will traverse this list and read the buildids,
stashing them at the end of the file and will set up a new
feature bit in the header bitmask.
'perf report' will then notice this feature and populate the
'dsos' list and set the build ids.
When reading the symtabs it will refuse to load from a file that
doesn't have the same build id. This improves the
reliability of the profiler output, as symbols and profiling
data is more guaranteed to match.
Example:
[root@doppio ~]# perf report | head
/home/acme/bin/perf with build id b1ea544ac3746e7538972548a09aadecc5753868 not found, continuing without symbols
# Samples: 2621434559
#
# Overhead Command Shared Object Symbol
# ........ ............... ............................. ......
#
7.91% init [kernel] [k] read_hpet
7.64% init [kernel] [k] mwait_idle_with_hints
7.60% swapper [kernel] [k] read_hpet
7.60% swapper [kernel] [k] mwait_idle_with_hints
3.65% init [kernel] [k] 0xffffffffa02339d9
[root@doppio ~]#
In this case the 'perf' binary was an older one, vanished,
so its symbols probably wouldn't match or would cause subtly
different (and misleading) output.
Next patches will support the kernel as well, reading the build
id notes for it and the modules from /sys.
Another patch should also introduce a new plumbing command:
'perf list-buildids'
that will then be used in porcelain that is distro specific to
fetch -debuginfo packages where such buildids are present. This
will in turn allow for one to run 'perf record' in one machine
and 'perf report' in another.
Future work on having the buildid sent directly from the kernel
in the PERF_RECORD_MMAP event is needed to close races, as the
DSO can be changed during a 'perf record' session, but this
patch at least helps with non-corner cases and current/older
kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.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: Jim Keniston <jkenisto@us.ibm.com>
Cc: K. Prasad <prasad@linux.vnet.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1257367843-26224-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-11-04 21:50:43 +01:00
|
|
|
write_event(buf, size);
|
2009-04-08 15:01:31 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
buf = &data[old & md->mask];
|
|
|
|
size = head - old;
|
|
|
|
old += size;
|
2009-06-03 19:27:19 +02:00
|
|
|
|
perf symbols: Use the buildids if present
With this change 'perf record' will intercept PERF_RECORD_MMAP
calls, creating a linked list of DSOs, then when the session
finishes, it will traverse this list and read the buildids,
stashing them at the end of the file and will set up a new
feature bit in the header bitmask.
'perf report' will then notice this feature and populate the
'dsos' list and set the build ids.
When reading the symtabs it will refuse to load from a file that
doesn't have the same build id. This improves the
reliability of the profiler output, as symbols and profiling
data is more guaranteed to match.
Example:
[root@doppio ~]# perf report | head
/home/acme/bin/perf with build id b1ea544ac3746e7538972548a09aadecc5753868 not found, continuing without symbols
# Samples: 2621434559
#
# Overhead Command Shared Object Symbol
# ........ ............... ............................. ......
#
7.91% init [kernel] [k] read_hpet
7.64% init [kernel] [k] mwait_idle_with_hints
7.60% swapper [kernel] [k] read_hpet
7.60% swapper [kernel] [k] mwait_idle_with_hints
3.65% init [kernel] [k] 0xffffffffa02339d9
[root@doppio ~]#
In this case the 'perf' binary was an older one, vanished,
so its symbols probably wouldn't match or would cause subtly
different (and misleading) output.
Next patches will support the kernel as well, reading the build
id notes for it and the modules from /sys.
Another patch should also introduce a new plumbing command:
'perf list-buildids'
that will then be used in porcelain that is distro specific to
fetch -debuginfo packages where such buildids are present. This
will in turn allow for one to run 'perf record' in one machine
and 'perf report' in another.
Future work on having the buildid sent directly from the kernel
in the PERF_RECORD_MMAP event is needed to close races, as the
DSO can be changed during a 'perf record' session, but this
patch at least helps with non-corner cases and current/older
kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.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: Jim Keniston <jkenisto@us.ibm.com>
Cc: K. Prasad <prasad@linux.vnet.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1257367843-26224-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-11-04 21:50:43 +01:00
|
|
|
write_event(buf, size);
|
2009-04-08 15:01:31 +02:00
|
|
|
|
|
|
|
md->prev = old;
|
2009-06-18 11:40:28 +02:00
|
|
|
mmap_write_tail(md, old);
|
2009-04-08 15:01:31 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static volatile int done = 0;
|
2009-06-10 15:55:59 +02:00
|
|
|
static volatile int signr = -1;
|
2009-04-08 15:01:31 +02:00
|
|
|
|
2009-05-05 17:50:27 +02:00
|
|
|
static void sig_handler(int sig)
|
2009-04-08 15:01:31 +02:00
|
|
|
{
|
2009-05-05 17:50:27 +02:00
|
|
|
done = 1;
|
2009-06-10 15:55:59 +02:00
|
|
|
signr = sig;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sig_atexit(void)
|
|
|
|
{
|
2009-10-04 02:35:01 +02:00
|
|
|
if (child_pid != -1)
|
|
|
|
kill(child_pid, SIGTERM);
|
|
|
|
|
2009-06-10 15:55:59 +02:00
|
|
|
if (signr == -1)
|
|
|
|
return;
|
|
|
|
|
|
|
|
signal(signr, SIG_DFL);
|
|
|
|
kill(getpid(), signr);
|
2009-04-08 15:01:31 +02:00
|
|
|
}
|
|
|
|
|
2009-06-05 13:18:41 +02:00
|
|
|
static int group_fd;
|
|
|
|
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int nr)
|
2009-06-25 17:05:54 +02:00
|
|
|
{
|
|
|
|
struct perf_header_attr *h_attr;
|
|
|
|
|
2009-12-12 00:24:02 +01:00
|
|
|
if (nr < session->header.attrs) {
|
|
|
|
h_attr = session->header.attr[nr];
|
2009-06-25 17:05:54 +02:00
|
|
|
} else {
|
|
|
|
h_attr = perf_header_attr__new(a);
|
2009-11-16 22:30:26 +01:00
|
|
|
if (h_attr != NULL)
|
2009-12-12 00:24:02 +01:00
|
|
|
if (perf_header__add_attr(&session->header, h_attr) < 0) {
|
2009-11-17 04:18:09 +01:00
|
|
|
perf_header_attr__delete(h_attr);
|
|
|
|
h_attr = NULL;
|
|
|
|
}
|
2009-06-25 17:05:54 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return h_attr;
|
|
|
|
}
|
|
|
|
|
2009-06-05 13:18:41 +02:00
|
|
|
static void create_counter(int counter, int cpu, pid_t pid)
|
2009-04-08 15:01:31 +02:00
|
|
|
{
|
2009-10-15 05:22:07 +02:00
|
|
|
char *filter = filters[counter];
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
struct perf_event_attr *attr = attrs + counter;
|
2009-06-25 17:05:54 +02:00
|
|
|
struct perf_header_attr *h_attr;
|
|
|
|
int track = !counter; /* only the first counter needs these */
|
2009-10-15 05:22:07 +02:00
|
|
|
int ret;
|
2009-06-25 17:05:54 +02:00
|
|
|
struct {
|
|
|
|
u64 count;
|
|
|
|
u64 time_enabled;
|
|
|
|
u64 time_running;
|
|
|
|
u64 id;
|
|
|
|
} read_data;
|
|
|
|
|
|
|
|
attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
|
|
|
|
PERF_FORMAT_TOTAL_TIME_RUNNING |
|
|
|
|
PERF_FORMAT_ID;
|
2009-05-05 17:50:27 +02:00
|
|
|
|
2009-08-13 10:27:18 +02:00
|
|
|
attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
|
2009-06-14 15:04:15 +02:00
|
|
|
|
2009-06-05 18:37:22 +02:00
|
|
|
if (freq) {
|
2009-06-10 21:45:22 +02:00
|
|
|
attr->sample_type |= PERF_SAMPLE_PERIOD;
|
2009-06-06 09:58:57 +02:00
|
|
|
attr->freq = 1;
|
|
|
|
attr->sample_freq = freq;
|
2009-06-05 18:37:22 +02:00
|
|
|
}
|
2009-06-14 15:04:15 +02:00
|
|
|
|
2009-06-24 21:12:48 +02:00
|
|
|
if (no_samples)
|
|
|
|
attr->sample_freq = 0;
|
|
|
|
|
|
|
|
if (inherit_stat)
|
|
|
|
attr->inherit_stat = 1;
|
|
|
|
|
2009-07-16 15:44:29 +02:00
|
|
|
if (sample_address)
|
|
|
|
attr->sample_type |= PERF_SAMPLE_ADDR;
|
|
|
|
|
2009-06-14 15:04:15 +02:00
|
|
|
if (call_graph)
|
|
|
|
attr->sample_type |= PERF_SAMPLE_CALLCHAIN;
|
|
|
|
|
2009-09-02 20:20:38 +02:00
|
|
|
if (raw_samples) {
|
2009-09-03 12:00:22 +02:00
|
|
|
attr->sample_type |= PERF_SAMPLE_TIME;
|
2009-08-13 10:27:19 +02:00
|
|
|
attr->sample_type |= PERF_SAMPLE_RAW;
|
2009-09-02 20:20:38 +02:00
|
|
|
attr->sample_type |= PERF_SAMPLE_CPU;
|
|
|
|
}
|
2009-08-07 01:25:54 +02:00
|
|
|
|
2009-06-06 09:58:57 +02:00
|
|
|
attr->mmap = track;
|
|
|
|
attr->comm = track;
|
|
|
|
attr->inherit = (cpu < 0) && inherit;
|
2009-06-10 15:03:06 +02:00
|
|
|
attr->disabled = 1;
|
2009-05-05 17:50:27 +02:00
|
|
|
|
2009-06-07 17:39:02 +02:00
|
|
|
try_again:
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
fd[nr_cpu][counter] = sys_perf_event_open(attr, pid, cpu, group_fd, 0);
|
2009-05-05 17:50:27 +02:00
|
|
|
|
2009-06-05 13:18:41 +02:00
|
|
|
if (fd[nr_cpu][counter] < 0) {
|
|
|
|
int err = errno;
|
2009-05-05 17:50:27 +02:00
|
|
|
|
2009-11-08 17:01:06 +01:00
|
|
|
if (err == EPERM || err == EACCES)
|
2009-06-07 17:39:02 +02:00
|
|
|
die("Permission error - are you root?\n");
|
2009-08-12 11:18:01 +02:00
|
|
|
else if (err == ENODEV && profile_cpu != -1)
|
|
|
|
die("No such device - did you specify an out-of-range profile CPU?\n");
|
2009-06-07 17:39:02 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If it's cycles then fall back to hrtimer
|
|
|
|
* based cpu-clock-tick sw counter, which
|
|
|
|
* is always available even if no PMU support:
|
|
|
|
*/
|
|
|
|
if (attr->type == PERF_TYPE_HARDWARE
|
2009-06-11 14:06:28 +02:00
|
|
|
&& attr->config == PERF_COUNT_HW_CPU_CYCLES) {
|
2009-06-07 17:39:02 +02:00
|
|
|
|
|
|
|
if (verbose)
|
|
|
|
warning(" ... trying to fall back to cpu-clock-ticks\n");
|
|
|
|
attr->type = PERF_TYPE_SOFTWARE;
|
2009-06-11 14:06:28 +02:00
|
|
|
attr->config = PERF_COUNT_SW_CPU_CLOCK;
|
2009-06-07 17:39:02 +02:00
|
|
|
goto try_again;
|
|
|
|
}
|
2009-06-07 17:46:24 +02:00
|
|
|
printf("\n");
|
|
|
|
error("perfcounter syscall returned with %d (%s)\n",
|
|
|
|
fd[nr_cpu][counter], strerror(err));
|
2009-11-16 06:25:53 +01:00
|
|
|
|
|
|
|
#if defined(__i386__) || defined(__x86_64__)
|
|
|
|
if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP)
|
|
|
|
die("No hardware sampling interrupt available. No APIC? If so then you can boot the kernel with the \"lapic\" boot parameter to force-enable it.\n");
|
|
|
|
#endif
|
|
|
|
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
|
2009-06-05 13:18:41 +02:00
|
|
|
exit(-1);
|
|
|
|
}
|
2009-06-07 17:39:02 +02:00
|
|
|
|
2009-06-25 17:05:54 +02:00
|
|
|
h_attr = get_header_attr(attr, counter);
|
2009-11-16 22:30:26 +01:00
|
|
|
if (h_attr == NULL)
|
|
|
|
die("nomem\n");
|
2009-06-25 17:05:54 +02:00
|
|
|
|
|
|
|
if (!file_new) {
|
|
|
|
if (memcmp(&h_attr->attr, attr, sizeof(*attr))) {
|
|
|
|
fprintf(stderr, "incompatible append\n");
|
|
|
|
exit(-1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-06-25 22:21:27 +02:00
|
|
|
if (read(fd[nr_cpu][counter], &read_data, sizeof(read_data)) == -1) {
|
|
|
|
perror("Unable to read perf file descriptor\n");
|
|
|
|
exit(-1);
|
|
|
|
}
|
2009-06-25 17:05:54 +02:00
|
|
|
|
2009-11-17 04:18:10 +01:00
|
|
|
if (perf_header_attr__add_id(h_attr, read_data.id) < 0) {
|
|
|
|
pr_warning("Not enough memory to add id\n");
|
|
|
|
exit(-1);
|
|
|
|
}
|
2009-06-25 17:05:54 +02:00
|
|
|
|
2009-06-05 13:18:41 +02:00
|
|
|
assert(fd[nr_cpu][counter] >= 0);
|
|
|
|
fcntl(fd[nr_cpu][counter], F_SETFL, O_NONBLOCK);
|
2009-05-05 17:50:27 +02:00
|
|
|
|
2009-06-05 13:18:41 +02:00
|
|
|
/*
|
|
|
|
* First counter acts as the group leader:
|
|
|
|
*/
|
|
|
|
if (group && group_fd == -1)
|
|
|
|
group_fd = fd[nr_cpu][counter];
|
2009-09-13 18:15:54 +02:00
|
|
|
if (multiplex && multiplex_fd == -1)
|
|
|
|
multiplex_fd = fd[nr_cpu][counter];
|
2009-06-05 13:18:41 +02:00
|
|
|
|
2009-09-13 18:15:54 +02:00
|
|
|
if (multiplex && fd[nr_cpu][counter] != multiplex_fd) {
|
2009-06-10 15:03:06 +02:00
|
|
|
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
ret = ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_SET_OUTPUT, multiplex_fd);
|
2009-09-13 18:15:54 +02:00
|
|
|
assert(ret != -1);
|
|
|
|
} else {
|
|
|
|
event_array[nr_poll].fd = fd[nr_cpu][counter];
|
|
|
|
event_array[nr_poll].events = POLLIN;
|
|
|
|
nr_poll++;
|
|
|
|
|
|
|
|
mmap_array[nr_cpu][counter].counter = counter;
|
|
|
|
mmap_array[nr_cpu][counter].prev = 0;
|
|
|
|
mmap_array[nr_cpu][counter].mask = mmap_pages*page_size - 1;
|
|
|
|
mmap_array[nr_cpu][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
|
|
|
|
PROT_READ|PROT_WRITE, MAP_SHARED, fd[nr_cpu][counter], 0);
|
|
|
|
if (mmap_array[nr_cpu][counter].base == MAP_FAILED) {
|
|
|
|
error("failed to mmap with %d (%s)\n", errno, strerror(errno));
|
|
|
|
exit(-1);
|
|
|
|
}
|
|
|
|
}
|
2009-09-14 08:57:15 +02:00
|
|
|
|
2009-10-15 05:22:07 +02:00
|
|
|
if (filter != NULL) {
|
|
|
|
ret = ioctl(fd[nr_cpu][counter],
|
|
|
|
PERF_EVENT_IOC_SET_FILTER, filter);
|
|
|
|
if (ret) {
|
|
|
|
error("failed to set filter with %d (%s)\n", errno,
|
|
|
|
strerror(errno));
|
|
|
|
exit(-1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_ENABLE);
|
2009-06-05 13:18:41 +02:00
|
|
|
}
|
2009-06-03 19:17:25 +02:00
|
|
|
|
2009-06-05 13:18:41 +02:00
|
|
|
static void open_counters(int cpu, pid_t pid)
|
|
|
|
{
|
|
|
|
int counter;
|
2009-05-05 17:50:27 +02:00
|
|
|
|
2009-06-05 13:18:41 +02:00
|
|
|
group_fd = -1;
|
|
|
|
for (counter = 0; counter < nr_counters; counter++)
|
|
|
|
create_counter(counter, cpu, pid);
|
|
|
|
|
2009-05-05 17:50:27 +02:00
|
|
|
nr_cpu++;
|
|
|
|
}
|
|
|
|
|
2009-06-18 23:22:55 +02:00
|
|
|
static void atexit_header(void)
|
|
|
|
{
|
2009-12-12 00:24:02 +01:00
|
|
|
session->header.data_size += bytes_written;
|
2009-06-18 23:22:55 +02:00
|
|
|
|
2009-12-12 00:24:02 +01:00
|
|
|
perf_header__write(&session->header, output, true);
|
2009-06-18 23:22:55 +02:00
|
|
|
}
|
|
|
|
|
2009-05-26 09:17:18 +02:00
|
|
|
static int __cmd_record(int argc, const char **argv)
|
2009-05-05 17:50:27 +02:00
|
|
|
{
|
|
|
|
int i, counter;
|
2009-06-02 22:59:57 +02:00
|
|
|
struct stat st;
|
2009-06-25 17:05:54 +02:00
|
|
|
pid_t pid = 0;
|
2009-06-02 22:59:57 +02:00
|
|
|
int flags;
|
2009-11-19 17:55:55 +01:00
|
|
|
int err;
|
2009-09-17 19:59:05 +02:00
|
|
|
unsigned long waking = 0;
|
2009-04-08 15:01:31 +02:00
|
|
|
|
|
|
|
page_size = sysconf(_SC_PAGE_SIZE);
|
|
|
|
nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
|
|
|
|
assert(nr_cpus <= MAX_NR_CPUS);
|
|
|
|
assert(nr_cpus >= 0);
|
|
|
|
|
2009-06-18 23:22:55 +02:00
|
|
|
atexit(sig_atexit);
|
|
|
|
signal(SIGCHLD, sig_handler);
|
|
|
|
signal(SIGINT, sig_handler);
|
|
|
|
|
2009-08-07 14:16:01 +02:00
|
|
|
if (!stat(output_name, &st) && st.st_size) {
|
|
|
|
if (!force && !append_file) {
|
|
|
|
fprintf(stderr, "Error, output file %s exists, use -A to append or -f to overwrite.\n",
|
|
|
|
output_name);
|
|
|
|
exit(-1);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
append_file = 0;
|
2009-06-02 15:52:24 +02:00
|
|
|
}
|
|
|
|
|
2009-06-02 22:59:57 +02:00
|
|
|
flags = O_CREAT|O_RDWR;
|
|
|
|
if (append_file)
|
2009-06-18 23:22:55 +02:00
|
|
|
file_new = 0;
|
2009-06-02 22:59:57 +02:00
|
|
|
else
|
|
|
|
flags |= O_TRUNC;
|
|
|
|
|
|
|
|
output = open(output_name, flags, S_IRUSR|S_IWUSR);
|
2009-04-08 15:01:31 +02:00
|
|
|
if (output < 0) {
|
|
|
|
perror("failed to create output file");
|
|
|
|
exit(-1);
|
|
|
|
}
|
|
|
|
|
2009-12-12 00:24:02 +01:00
|
|
|
session = perf_session__new(output_name, O_WRONLY, force);
|
|
|
|
if (session == NULL) {
|
2009-11-17 04:18:11 +01:00
|
|
|
pr_err("Not enough memory for reading perf file header\n");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2009-11-19 17:55:55 +01:00
|
|
|
if (!file_new) {
|
2009-12-12 00:24:02 +01:00
|
|
|
err = perf_header__read(&session->header, output);
|
2009-11-19 17:55:55 +01:00
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2009-08-17 23:07:50 +02:00
|
|
|
if (raw_samples) {
|
2009-12-12 00:24:02 +01:00
|
|
|
perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
|
2009-08-17 23:07:50 +02:00
|
|
|
} else {
|
|
|
|
for (i = 0; i < nr_counters; i++) {
|
|
|
|
if (attrs[i].sample_type & PERF_SAMPLE_RAW) {
|
2009-12-12 00:24:02 +01:00
|
|
|
perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
|
2009-08-17 23:07:50 +02:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2009-10-06 23:36:47 +02:00
|
|
|
|
2009-06-18 23:22:55 +02:00
|
|
|
atexit(atexit_header);
|
|
|
|
|
2009-05-15 03:50:46 +02:00
|
|
|
if (!system_wide) {
|
2009-06-25 17:05:54 +02:00
|
|
|
pid = target_pid;
|
|
|
|
if (pid == -1)
|
|
|
|
pid = getpid();
|
|
|
|
|
2009-08-12 11:18:01 +02:00
|
|
|
open_counters(profile_cpu, pid);
|
|
|
|
} else {
|
|
|
|
if (profile_cpu != -1) {
|
|
|
|
open_counters(profile_cpu, target_pid);
|
|
|
|
} else {
|
|
|
|
for (i = 0; i < nr_cpus; i++)
|
|
|
|
open_counters(i, target_pid);
|
|
|
|
}
|
|
|
|
}
|
2009-04-08 15:01:31 +02:00
|
|
|
|
2009-11-19 17:55:56 +01:00
|
|
|
if (file_new) {
|
2009-12-12 00:24:02 +01:00
|
|
|
err = perf_header__write(&session->header, output, false);
|
2009-11-19 17:55:56 +01:00
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
}
|
2009-06-25 17:05:54 +02:00
|
|
|
|
2009-10-26 22:23:18 +01:00
|
|
|
if (!system_wide)
|
|
|
|
event__synthesize_thread(pid, process_synthesized_event);
|
|
|
|
else
|
|
|
|
event__synthesize_threads(process_synthesized_event);
|
2009-06-25 17:05:54 +02:00
|
|
|
|
2009-05-27 10:10:51 +02:00
|
|
|
if (target_pid == -1 && argc) {
|
2009-05-15 03:50:46 +02:00
|
|
|
pid = fork();
|
|
|
|
if (pid < 0)
|
2009-11-11 04:51:02 +01:00
|
|
|
die("failed to fork");
|
2009-04-08 15:01:31 +02:00
|
|
|
|
2009-05-15 03:50:46 +02:00
|
|
|
if (!pid) {
|
2009-05-26 09:17:18 +02:00
|
|
|
if (execvp(argv[0], (char **)argv)) {
|
2009-05-15 03:50:46 +02:00
|
|
|
perror(argv[0]);
|
|
|
|
exit(-1);
|
|
|
|
}
|
2009-11-11 04:51:02 +01:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Wait a bit for the execv'ed child to appear
|
|
|
|
* and be updated in /proc
|
|
|
|
* FIXME: Do you know a less heuristical solution?
|
|
|
|
*/
|
|
|
|
usleep(1000);
|
|
|
|
event__synthesize_thread(pid,
|
|
|
|
process_synthesized_event);
|
2009-04-08 15:01:31 +02:00
|
|
|
}
|
2009-10-04 02:35:01 +02:00
|
|
|
|
|
|
|
child_pid = pid;
|
2009-04-08 15:01:31 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (realtime_prio) {
|
|
|
|
struct sched_param param;
|
|
|
|
|
|
|
|
param.sched_priority = realtime_prio;
|
|
|
|
if (sched_setscheduler(0, SCHED_FIFO, ¶m)) {
|
2009-10-21 21:34:06 +02:00
|
|
|
pr_err("Could not set realtime priority.\n");
|
2009-04-08 15:01:31 +02:00
|
|
|
exit(-1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-06-24 21:12:48 +02:00
|
|
|
for (;;) {
|
2009-06-05 14:29:10 +02:00
|
|
|
int hits = samples;
|
2009-04-08 15:01:31 +02:00
|
|
|
|
2009-05-05 17:50:27 +02:00
|
|
|
for (i = 0; i < nr_cpu; i++) {
|
2009-09-13 18:15:54 +02:00
|
|
|
for (counter = 0; counter < nr_counters; counter++) {
|
|
|
|
if (mmap_array[i][counter].base)
|
|
|
|
mmap_read(&mmap_array[i][counter]);
|
|
|
|
}
|
2009-04-08 15:01:31 +02:00
|
|
|
}
|
|
|
|
|
2009-06-24 21:12:48 +02:00
|
|
|
if (hits == samples) {
|
|
|
|
if (done)
|
|
|
|
break;
|
2009-11-19 17:55:55 +01:00
|
|
|
err = poll(event_array, nr_poll, -1);
|
2009-09-17 19:59:05 +02:00
|
|
|
waking++;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (done) {
|
|
|
|
for (i = 0; i < nr_cpu; i++) {
|
|
|
|
for (counter = 0; counter < nr_counters; counter++)
|
perf: Do the big rename: Performance Counters -> Performance Events
Bye-bye Performance Counters, welcome Performance Events!
In the past few months the perfcounters subsystem has grown out its
initial role of counting hardware events, and has become (and is
becoming) a much broader generic event enumeration, reporting, logging,
monitoring, analysis facility.
Naming its core object 'perf_counter' and naming the subsystem
'perfcounters' has become more and more of a misnomer. With pending
code like hw-breakpoints support the 'counter' name is less and
less appropriate.
All in one, we've decided to rename the subsystem to 'performance
events' and to propagate this rename through all fields, variables
and API names. (in an ABI compatible fashion)
The word 'event' is also a bit shorter than 'counter' - which makes
it slightly more convenient to write/handle as well.
Thanks goes to Stephane Eranian who first observed this misnomer and
suggested a rename.
User-space tooling and ABI compatibility is not affected - this patch
should be function-invariant. (Also, defconfigs were not touched to
keep the size down.)
This patch has been generated via the following script:
FILES=$(find * -type f | grep -vE 'oprofile|[^K]config')
sed -i \
-e 's/PERF_EVENT_/PERF_RECORD_/g' \
-e 's/PERF_COUNTER/PERF_EVENT/g' \
-e 's/perf_counter/perf_event/g' \
-e 's/nb_counters/nb_events/g' \
-e 's/swcounter/swevent/g' \
-e 's/tpcounter_event/tp_event/g' \
$FILES
for N in $(find . -name perf_counter.[ch]); do
M=$(echo $N | sed 's/perf_counter/perf_event/g')
mv $N $M
done
FILES=$(find . -name perf_event.*)
sed -i \
-e 's/COUNTER_MASK/REG_MASK/g' \
-e 's/COUNTER/EVENT/g' \
-e 's/\<event\>/event_id/g' \
-e 's/counter/event/g' \
-e 's/Counter/Event/g' \
$FILES
... to keep it as correct as possible. This script can also be
used by anyone who has pending perfcounters patches - it converts
a Linux kernel tree over to the new naming. We tried to time this
change to the point in time where the amount of pending patches
is the smallest: the end of the merge window.
Namespace clashes were fixed up in a preparatory patch - and some
stylistic fallout will be fixed up in a subsequent patch.
( NOTE: 'counters' are still the proper terminology when we deal
with hardware registers - and these sed scripts are a bit
over-eager in renaming them. I've undone some of that, but
in case there's something left where 'counter' would be
better than 'event' we can undo that on an individual basis
instead of touching an otherwise nicely automated patch. )
Suggested-by: Stephane Eranian <eranian@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul Mackerras <paulus@samba.org>
Reviewed-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <linux-arch@vger.kernel.org>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-09-21 12:02:48 +02:00
|
|
|
ioctl(fd[i][counter], PERF_EVENT_IOC_DISABLE);
|
2009-09-17 19:59:05 +02:00
|
|
|
}
|
2009-06-24 21:12:48 +02:00
|
|
|
}
|
2009-04-08 15:01:31 +02:00
|
|
|
}
|
|
|
|
|
2009-09-17 19:59:05 +02:00
|
|
|
fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);
|
|
|
|
|
2009-06-03 19:27:19 +02:00
|
|
|
/*
|
|
|
|
* Approximate RIP event size: 24 bytes.
|
|
|
|
*/
|
|
|
|
fprintf(stderr,
|
2009-06-05 14:29:10 +02:00
|
|
|
"[ perf record: Captured and wrote %.3f MB %s (~%lld samples) ]\n",
|
2009-06-03 19:27:19 +02:00
|
|
|
(double)bytes_written / 1024.0 / 1024.0,
|
|
|
|
output_name,
|
|
|
|
bytes_written / 24);
|
2009-06-02 23:43:11 +02:00
|
|
|
|
2009-04-08 15:01:31 +02:00
|
|
|
return 0;
|
|
|
|
}
|
2009-05-26 09:17:18 +02:00
|
|
|
|
|
|
|
static const char * const record_usage[] = {
|
2009-05-28 16:25:34 +02:00
|
|
|
"perf record [<options>] [<command>]",
|
|
|
|
"perf record [<options>] -- <command> [<options>]",
|
2009-05-26 09:17:18 +02:00
|
|
|
NULL
|
|
|
|
};
|
|
|
|
|
2009-05-26 09:17:18 +02:00
|
|
|
static const struct option options[] = {
|
2009-05-26 09:17:18 +02:00
|
|
|
OPT_CALLBACK('e', "event", NULL, "event",
|
2009-06-06 12:24:17 +02:00
|
|
|
"event selector. use 'perf list' to list available events",
|
|
|
|
parse_events),
|
2009-10-15 05:22:07 +02:00
|
|
|
OPT_CALLBACK(0, "filter", NULL, "filter",
|
|
|
|
"event filter", parse_filter),
|
2009-05-26 09:17:18 +02:00
|
|
|
OPT_INTEGER('p', "pid", &target_pid,
|
|
|
|
"record events on existing pid"),
|
|
|
|
OPT_INTEGER('r', "realtime", &realtime_prio,
|
|
|
|
"collect data with this RT SCHED_FIFO priority"),
|
2009-08-13 10:27:19 +02:00
|
|
|
OPT_BOOLEAN('R', "raw-samples", &raw_samples,
|
|
|
|
"collect raw sample records from all opened counters"),
|
2009-05-26 09:17:18 +02:00
|
|
|
OPT_BOOLEAN('a', "all-cpus", &system_wide,
|
|
|
|
"system-wide collection from all CPUs"),
|
2009-06-02 22:59:57 +02:00
|
|
|
OPT_BOOLEAN('A', "append", &append_file,
|
|
|
|
"append to the output file to do incremental profiling"),
|
2009-08-12 11:18:01 +02:00
|
|
|
OPT_INTEGER('C', "profile_cpu", &profile_cpu,
|
|
|
|
"CPU to profile on"),
|
2009-06-02 15:52:24 +02:00
|
|
|
OPT_BOOLEAN('f', "force", &force,
|
|
|
|
"overwrite existing data file"),
|
2009-06-03 11:24:33 +02:00
|
|
|
OPT_LONG('c', "count", &default_interval,
|
2009-06-02 22:59:57 +02:00
|
|
|
"event period to sample"),
|
|
|
|
OPT_STRING('o', "output", &output_name, "file",
|
|
|
|
"output file name"),
|
|
|
|
OPT_BOOLEAN('i', "inherit", &inherit,
|
|
|
|
"child tasks inherit counters"),
|
2009-06-05 13:27:02 +02:00
|
|
|
OPT_INTEGER('F', "freq", &freq,
|
|
|
|
"profile at this frequency"),
|
2009-06-02 22:59:57 +02:00
|
|
|
OPT_INTEGER('m', "mmap-pages", &mmap_pages,
|
|
|
|
"number of mmap data pages"),
|
2009-06-14 15:04:15 +02:00
|
|
|
OPT_BOOLEAN('g', "call-graph", &call_graph,
|
|
|
|
"do call-graph (stack chain/backtrace) recording"),
|
2009-06-07 17:39:02 +02:00
|
|
|
OPT_BOOLEAN('v', "verbose", &verbose,
|
|
|
|
"be more verbose (show counter open errors, etc)"),
|
2009-06-24 21:12:48 +02:00
|
|
|
OPT_BOOLEAN('s', "stat", &inherit_stat,
|
|
|
|
"per thread counts"),
|
2009-07-16 15:44:29 +02:00
|
|
|
OPT_BOOLEAN('d', "data", &sample_address,
|
|
|
|
"Sample addresses"),
|
2009-06-24 21:12:48 +02:00
|
|
|
OPT_BOOLEAN('n', "no-samples", &no_samples,
|
|
|
|
"don't sample"),
|
2009-09-14 08:57:15 +02:00
|
|
|
OPT_BOOLEAN('M', "multiplex", &multiplex,
|
|
|
|
"multiplex counter output in a single channel"),
|
2009-05-26 09:17:18 +02:00
|
|
|
OPT_END()
|
|
|
|
};
|
|
|
|
|
2009-07-01 12:37:06 +02:00
|
|
|
int cmd_record(int argc, const char **argv, const char *prefix __used)
|
2009-05-26 09:17:18 +02:00
|
|
|
{
|
|
|
|
int counter;
|
|
|
|
|
perf symbols: Use the buildids if present
With this change 'perf record' will intercept PERF_RECORD_MMAP
calls, creating a linked list of DSOs, then when the session
finishes, it will traverse this list and read the buildids,
stashing them at the end of the file and will set up a new
feature bit in the header bitmask.
'perf report' will then notice this feature and populate the
'dsos' list and set the build ids.
When reading the symtabs it will refuse to load from a file that
doesn't have the same build id. This improves the
reliability of the profiler output, as symbols and profiling
data is more guaranteed to match.
Example:
[root@doppio ~]# perf report | head
/home/acme/bin/perf with build id b1ea544ac3746e7538972548a09aadecc5753868 not found, continuing without symbols
# Samples: 2621434559
#
# Overhead Command Shared Object Symbol
# ........ ............... ............................. ......
#
7.91% init [kernel] [k] read_hpet
7.64% init [kernel] [k] mwait_idle_with_hints
7.60% swapper [kernel] [k] read_hpet
7.60% swapper [kernel] [k] mwait_idle_with_hints
3.65% init [kernel] [k] 0xffffffffa02339d9
[root@doppio ~]#
In this case the 'perf' binary was an older one, vanished,
so its symbols probably wouldn't match or would cause subtly
different (and misleading) output.
Next patches will support the kernel as well, reading the build
id notes for it and the modules from /sys.
Another patch should also introduce a new plumbing command:
'perf list-buildids'
that will then be used in porcelain that is distro specific to
fetch -debuginfo packages where such buildids are present. This
will in turn allow for one to run 'perf record' in one machine
and 'perf report' in another.
Future work on having the buildid sent directly from the kernel
in the PERF_RECORD_MMAP event is needed to close races, as the
DSO can be changed during a 'perf record' session, but this
patch at least helps with non-corner cases and current/older
kernels.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.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: Jim Keniston <jkenisto@us.ibm.com>
Cc: K. Prasad <prasad@linux.vnet.ibm.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1257367843-26224-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-11-04 21:50:43 +01:00
|
|
|
symbol__init(0);
|
|
|
|
|
2009-07-22 15:04:12 +02:00
|
|
|
argc = parse_options(argc, argv, options, record_usage,
|
|
|
|
PARSE_OPT_STOP_AT_NON_OPTION);
|
2009-05-27 10:10:51 +02:00
|
|
|
if (!argc && target_pid == -1 && !system_wide)
|
2009-05-26 09:17:18 +02:00
|
|
|
usage_with_options(record_usage, options);
|
|
|
|
|
2009-06-11 23:11:50 +02:00
|
|
|
if (!nr_counters) {
|
|
|
|
nr_counters = 1;
|
|
|
|
attrs[0].type = PERF_TYPE_HARDWARE;
|
|
|
|
attrs[0].config = PERF_COUNT_HW_CPU_CYCLES;
|
|
|
|
}
|
2009-05-26 09:17:18 +02:00
|
|
|
|
2009-10-12 07:56:03 +02:00
|
|
|
/*
|
|
|
|
* User specified count overrides default frequency.
|
|
|
|
*/
|
|
|
|
if (default_interval)
|
|
|
|
freq = 0;
|
|
|
|
else if (freq) {
|
|
|
|
default_interval = freq;
|
|
|
|
} else {
|
|
|
|
fprintf(stderr, "frequency and count are zero, aborting\n");
|
|
|
|
exit(EXIT_FAILURE);
|
|
|
|
}
|
|
|
|
|
2009-05-26 09:17:18 +02:00
|
|
|
for (counter = 0; counter < nr_counters; counter++) {
|
2009-06-06 09:58:57 +02:00
|
|
|
if (attrs[counter].sample_period)
|
2009-05-26 09:17:18 +02:00
|
|
|
continue;
|
|
|
|
|
2009-06-06 09:58:57 +02:00
|
|
|
attrs[counter].sample_period = default_interval;
|
2009-05-26 09:17:18 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return __cmd_record(argc, argv);
|
|
|
|
}
|