OpenCloudOS-Kernel/tools/perf/util/session.h

169 lines
4.5 KiB
C
Raw Normal View History

#ifndef __PERF_SESSION_H
#define __PERF_SESSION_H
2010-05-11 00:04:11 +08:00
#include "hist.h"
#include "event.h"
#include "header.h"
#include "symbol.h"
#include "thread.h"
#include <linux/rbtree.h>
#include "../../../include/linux/perf_event.h"
perf: Generalize perf lock's sample event reordering to the session layer The sample events recorded by perf record are not time ordered because we have one buffer per cpu for each event (even demultiplexed per task/per cpu for task bound events). But when we read trace events we want them to be ordered by time because many state machines are involved. There are currently two ways perf tools deal with that: - use -M to multiplex every buffers (perf sched, perf kmem) But this creates a lot of contention in SMP machines on record time. - use a post-processing time reordering (perf timechart, perf lock) The reordering used by timechart is simple but doesn't scale well with huge flow of events, in terms of performance and memory use (unusable with perf lock for example). Perf lock has its own samples reordering that flushes its memory use in a regular basis and that uses a sorting based on the previous event queued (a new event to be queued is close to the previous one most of the time). This patch proposes to export perf lock's samples reordering facility to the session layer that reads the events. So if a tool wants to get ordered sample events, it needs to set its struct perf_event_ops::ordered_samples to true and that's it. This prepares tracing based perf tools to get rid of the need to use buffers multiplexing (-M) or to implement their own reordering. Also lower the flush period to 2 as it's sufficient already. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Masami Hiramatsu <mhiramat@redhat.com> Cc: Tom Zanussi <tzanussi@gmail.com>
2010-04-24 06:04:12 +08:00
struct sample_queue;
struct ip_callchain;
struct thread;
perf: Generalize perf lock's sample event reordering to the session layer The sample events recorded by perf record are not time ordered because we have one buffer per cpu for each event (even demultiplexed per task/per cpu for task bound events). But when we read trace events we want them to be ordered by time because many state machines are involved. There are currently two ways perf tools deal with that: - use -M to multiplex every buffers (perf sched, perf kmem) But this creates a lot of contention in SMP machines on record time. - use a post-processing time reordering (perf timechart, perf lock) The reordering used by timechart is simple but doesn't scale well with huge flow of events, in terms of performance and memory use (unusable with perf lock for example). Perf lock has its own samples reordering that flushes its memory use in a regular basis and that uses a sorting based on the previous event queued (a new event to be queued is close to the previous one most of the time). This patch proposes to export perf lock's samples reordering facility to the session layer that reads the events. So if a tool wants to get ordered sample events, it needs to set its struct perf_event_ops::ordered_samples to true and that's it. This prepares tracing based perf tools to get rid of the need to use buffers multiplexing (-M) or to implement their own reordering. Also lower the flush period to 2 as it's sufficient already. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Masami Hiramatsu <mhiramat@redhat.com> Cc: Tom Zanussi <tzanussi@gmail.com>
2010-04-24 06:04:12 +08:00
struct ordered_samples {
u64 last_flush;
perf: Provide a new deterministic events reordering algorithm The current events reordering algorithm is based on a heuristic that gets broken once we deal with a very fast flow of events. Indeed the time period based flushing is not suitable anymore in the following case, assuming we have a flush period of two seconds. CPU 0 | CPU 1 | cnt1 timestamps | cnt1 timestamps | 0 | 0 1 | 1 2 | 2 3 | 3 [...] | [...] 4 seconds later If we spend too much time to read the buffers (case of a lot of events to record in each buffers or when we have a lot of CPU buffers to read), in the next pass the CPU 0 buffer could contain a slice of several seconds of events. We'll read them all and notice we've reached the period to flush. In the above example we flush the first half of the CPU 0 buffer, then we read the CPU 1 buffer where we have events that were on the flush slice and then the reordering fails. It's simple to reproduce with: perf lock record perf bench sched messaging To solve this, we use a new solution that doesn't rely on an heuristical time slice period anymore but on a deterministic basis based on how perf record does its job. perf record saves the buffers through passes. A pass is a tour on every buffers from every CPUs. This is made in order: for each CPU we read the buffers of every counters. So the more buffers we visit, the later will be the timstamps of their events. When perf record finishes a pass it records a PERF_RECORD_FINISHED_ROUND pseudo event. We record the max timestamp t found in the pass n. Assuming these timestamps are monotonic across cpus, we know that if a buffer still has events with timestamps below t, they will be all available and then read in the pass n + 1. Hence when we start to read the pass n + 2, we can safely flush every events with timestamps below t. ============ PASS n ================= CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 1 | 2 2 | 3 - | 4 <--- max recorded ============ PASS n + 1 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 3 | 5 4 | 6 5 | 7 <---- max recorded Flush every events below timestamp 4 ============ PASS n + 2 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 6 | 8 7 | 9 - | 10 Flush every events below timestamp 7 etc... It also works on perf.data versions that don't have PERF_RECORD_FINISHED_ROUND pseudo events. The difference is that the events will be only flushed in the end of the perf.data processing. It will then consume more memory and scale less with large perf.data files. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Tom Zanussi <tzanussi@gmail.com> Cc: Masami Hiramatsu <mhiramat@redhat.com>
2010-05-03 21:14:33 +08:00
u64 next_flush;
u64 max_timestamp;
struct list_head samples;
struct list_head sample_cache;
struct list_head to_free;
struct sample_queue *sample_buffer;
struct sample_queue *last_sample;
int sample_buffer_idx;
perf: Generalize perf lock's sample event reordering to the session layer The sample events recorded by perf record are not time ordered because we have one buffer per cpu for each event (even demultiplexed per task/per cpu for task bound events). But when we read trace events we want them to be ordered by time because many state machines are involved. There are currently two ways perf tools deal with that: - use -M to multiplex every buffers (perf sched, perf kmem) But this creates a lot of contention in SMP machines on record time. - use a post-processing time reordering (perf timechart, perf lock) The reordering used by timechart is simple but doesn't scale well with huge flow of events, in terms of performance and memory use (unusable with perf lock for example). Perf lock has its own samples reordering that flushes its memory use in a regular basis and that uses a sorting based on the previous event queued (a new event to be queued is close to the previous one most of the time). This patch proposes to export perf lock's samples reordering facility to the session layer that reads the events. So if a tool wants to get ordered sample events, it needs to set its struct perf_event_ops::ordered_samples to true and that's it. This prepares tracing based perf tools to get rid of the need to use buffers multiplexing (-M) or to implement their own reordering. Also lower the flush period to 2 as it's sufficient already. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Masami Hiramatsu <mhiramat@redhat.com> Cc: Tom Zanussi <tzanussi@gmail.com>
2010-04-24 06:04:12 +08:00
};
struct perf_session {
struct perf_header header;
unsigned long size;
unsigned long mmap_window;
struct rb_root threads;
struct list_head dead_threads;
struct thread *last_match;
struct machine host_machine;
struct rb_root machines;
2010-05-11 00:04:11 +08:00
struct rb_root hists_tree;
/*
* FIXME: should point to the first entry in hists_tree and
* be a hists instance. Right now its only 'report'
* that is using ->hists_tree while all the rest use
* ->hists.
*/
struct hists hists;
u64 sample_type;
int fd;
bool fd_pipe;
perf: add perf-inject builtin Currently, perf 'live mode' writes build-ids at the end of the session, which isn't actually useful for processing live mode events. What would be better would be to have the build-ids sent before any of the samples that reference them, which can be done by processing the event stream and retrieving the build-ids on the first hit. Doing that in perf-record itself, however, is off-limits. This patch introduces perf-inject, which does the same job while leaving perf-record untouched. Normal mode perf still records the build-ids at the end of the session as it should, but for live mode, perf-inject can be injected in between the record and report steps e.g.: perf record -o - ./hackbench 10 | perf inject -v -b | perf report -v -i - perf-inject reads a perf-record event stream and repipes it to stdout. At any point the processing code can inject other events into the event stream - in this case build-ids (-b option) are read and injected as needed into the event stream. Build-ids are just the first user of perf-inject - potentially anything that needs userspace processing to augment the trace stream with additional information could make use of this facility. Cc: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frédéric Weisbecker <fweisbec@gmail.com> LKML-Reference: <1272696080-16435-3-git-send-email-tzanussi@gmail.com> Signed-off-by: Tom Zanussi <tzanussi@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-05-01 14:41:20 +08:00
bool repipe;
perf tools: Ask for ID PERF_SAMPLE_ info on all PERF_RECORD_ events So that we can use -T == --timestamp, asking for PERF_SAMPLE_TIME: $ perf record -aT $ perf report -D | grep PERF_RECORD_ <SNIP> 3 5951915425 0x47530 [0x58]: PERF_RECORD_SAMPLE(IP, 1): 16811/16811: 0xffffffff8138c1a2 period: 215979 cpu:3 3 5952026879 0x47588 [0x90]: PERF_RECORD_SAMPLE(IP, 1): 16811/16811: 0xffffffff810cb480 period: 215979 cpu:3 3 5952059959 0x47618 [0x38]: PERF_RECORD_FORK(6853:6853):(16811:16811) 3 5952138878 0x47650 [0x78]: PERF_RECORD_SAMPLE(IP, 1): 16811/16811: 0xffffffff811bac35 period: 431478 cpu:3 3 5952375068 0x476c8 [0x30]: PERF_RECORD_COMM: find:6853 3 5952395923 0x476f8 [0x50]: PERF_RECORD_MMAP 6853/6853: [0x400000(0x25000) @ 0]: /usr/bin/find 3 5952413756 0x47748 [0xa0]: PERF_RECORD_SAMPLE(IP, 1): 6853/6853: 0xffffffff810d080f period: 859332 cpu:3 3 5952419837 0x477e8 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f44600000(0x21d000) @ 0]: /lib64/ld-2.5.so 3 5952437929 0x47840 [0x48]: PERF_RECORD_MMAP 6853/6853: [0x7fff7e1c9000(0x1000) @ 0x7fff7e1c9000]: [vdso] 3 5952570127 0x47888 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f46200000(0x218000) @ 0]: /lib64/libselinux.so.1 3 5952623637 0x478e0 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f44a00000(0x356000) @ 0]: /lib64/libc-2.5.so 3 5952675720 0x47938 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f44e00000(0x204000) @ 0]: /lib64/libdl-2.5.so 3 5952710080 0x47990 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f45a00000(0x246000) @ 0]: /lib64/libsepol.so.1 3 5952847802 0x479e8 [0x58]: PERF_RECORD_SAMPLE(IP, 1): 6853/6853: 0xffffffff813897f0 period: 1142536 cpu:3 <SNIP> First column is the cpu and the second the timestamp. That way we can investigate problems in the event stream. If the new perf binary is run on an older kernel, it will disable this feature automatically. Tested-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ian Munsie <imunsie@au1.ibm.com> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: Ian Munsie <imunsie@au1.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Stephane Eranian <eranian@google.com> LKML-Reference: <1291318772-30880-5-git-send-email-acme@infradead.org> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-12-02 20:25:28 +08:00
bool sample_id_all;
u16 id_hdr_size;
int cwdlen;
char *cwd;
perf: Generalize perf lock's sample event reordering to the session layer The sample events recorded by perf record are not time ordered because we have one buffer per cpu for each event (even demultiplexed per task/per cpu for task bound events). But when we read trace events we want them to be ordered by time because many state machines are involved. There are currently two ways perf tools deal with that: - use -M to multiplex every buffers (perf sched, perf kmem) But this creates a lot of contention in SMP machines on record time. - use a post-processing time reordering (perf timechart, perf lock) The reordering used by timechart is simple but doesn't scale well with huge flow of events, in terms of performance and memory use (unusable with perf lock for example). Perf lock has its own samples reordering that flushes its memory use in a regular basis and that uses a sorting based on the previous event queued (a new event to be queued is close to the previous one most of the time). This patch proposes to export perf lock's samples reordering facility to the session layer that reads the events. So if a tool wants to get ordered sample events, it needs to set its struct perf_event_ops::ordered_samples to true and that's it. This prepares tracing based perf tools to get rid of the need to use buffers multiplexing (-M) or to implement their own reordering. Also lower the flush period to 2 as it's sufficient already. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@elte.hu> Cc: Masami Hiramatsu <mhiramat@redhat.com> Cc: Tom Zanussi <tzanussi@gmail.com>
2010-04-24 06:04:12 +08:00
struct ordered_samples ordered_samples;
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
struct callchain_cursor callchain_cursor;
char filename[0];
};
perf: Provide a new deterministic events reordering algorithm The current events reordering algorithm is based on a heuristic that gets broken once we deal with a very fast flow of events. Indeed the time period based flushing is not suitable anymore in the following case, assuming we have a flush period of two seconds. CPU 0 | CPU 1 | cnt1 timestamps | cnt1 timestamps | 0 | 0 1 | 1 2 | 2 3 | 3 [...] | [...] 4 seconds later If we spend too much time to read the buffers (case of a lot of events to record in each buffers or when we have a lot of CPU buffers to read), in the next pass the CPU 0 buffer could contain a slice of several seconds of events. We'll read them all and notice we've reached the period to flush. In the above example we flush the first half of the CPU 0 buffer, then we read the CPU 1 buffer where we have events that were on the flush slice and then the reordering fails. It's simple to reproduce with: perf lock record perf bench sched messaging To solve this, we use a new solution that doesn't rely on an heuristical time slice period anymore but on a deterministic basis based on how perf record does its job. perf record saves the buffers through passes. A pass is a tour on every buffers from every CPUs. This is made in order: for each CPU we read the buffers of every counters. So the more buffers we visit, the later will be the timstamps of their events. When perf record finishes a pass it records a PERF_RECORD_FINISHED_ROUND pseudo event. We record the max timestamp t found in the pass n. Assuming these timestamps are monotonic across cpus, we know that if a buffer still has events with timestamps below t, they will be all available and then read in the pass n + 1. Hence when we start to read the pass n + 2, we can safely flush every events with timestamps below t. ============ PASS n ================= CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 1 | 2 2 | 3 - | 4 <--- max recorded ============ PASS n + 1 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 3 | 5 4 | 6 5 | 7 <---- max recorded Flush every events below timestamp 4 ============ PASS n + 2 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 6 | 8 7 | 9 - | 10 Flush every events below timestamp 7 etc... It also works on perf.data versions that don't have PERF_RECORD_FINISHED_ROUND pseudo events. The difference is that the events will be only flushed in the end of the perf.data processing. It will then consume more memory and scale less with large perf.data files. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Tom Zanussi <tzanussi@gmail.com> Cc: Masami Hiramatsu <mhiramat@redhat.com>
2010-05-03 21:14:33 +08:00
struct perf_event_ops;
typedef int (*event_op)(union perf_event *self, struct perf_sample *sample,
struct perf_session *session);
typedef int (*event_synth_op)(union perf_event *self,
struct perf_session *session);
typedef int (*event_op2)(union perf_event *self, struct perf_session *session,
perf: Provide a new deterministic events reordering algorithm The current events reordering algorithm is based on a heuristic that gets broken once we deal with a very fast flow of events. Indeed the time period based flushing is not suitable anymore in the following case, assuming we have a flush period of two seconds. CPU 0 | CPU 1 | cnt1 timestamps | cnt1 timestamps | 0 | 0 1 | 1 2 | 2 3 | 3 [...] | [...] 4 seconds later If we spend too much time to read the buffers (case of a lot of events to record in each buffers or when we have a lot of CPU buffers to read), in the next pass the CPU 0 buffer could contain a slice of several seconds of events. We'll read them all and notice we've reached the period to flush. In the above example we flush the first half of the CPU 0 buffer, then we read the CPU 1 buffer where we have events that were on the flush slice and then the reordering fails. It's simple to reproduce with: perf lock record perf bench sched messaging To solve this, we use a new solution that doesn't rely on an heuristical time slice period anymore but on a deterministic basis based on how perf record does its job. perf record saves the buffers through passes. A pass is a tour on every buffers from every CPUs. This is made in order: for each CPU we read the buffers of every counters. So the more buffers we visit, the later will be the timstamps of their events. When perf record finishes a pass it records a PERF_RECORD_FINISHED_ROUND pseudo event. We record the max timestamp t found in the pass n. Assuming these timestamps are monotonic across cpus, we know that if a buffer still has events with timestamps below t, they will be all available and then read in the pass n + 1. Hence when we start to read the pass n + 2, we can safely flush every events with timestamps below t. ============ PASS n ================= CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 1 | 2 2 | 3 - | 4 <--- max recorded ============ PASS n + 1 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 3 | 5 4 | 6 5 | 7 <---- max recorded Flush every events below timestamp 4 ============ PASS n + 2 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 6 | 8 7 | 9 - | 10 Flush every events below timestamp 7 etc... It also works on perf.data versions that don't have PERF_RECORD_FINISHED_ROUND pseudo events. The difference is that the events will be only flushed in the end of the perf.data processing. It will then consume more memory and scale less with large perf.data files. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Tom Zanussi <tzanussi@gmail.com> Cc: Masami Hiramatsu <mhiramat@redhat.com>
2010-05-03 21:14:33 +08:00
struct perf_event_ops *ops);
struct perf_event_ops {
perf: Provide a new deterministic events reordering algorithm The current events reordering algorithm is based on a heuristic that gets broken once we deal with a very fast flow of events. Indeed the time period based flushing is not suitable anymore in the following case, assuming we have a flush period of two seconds. CPU 0 | CPU 1 | cnt1 timestamps | cnt1 timestamps | 0 | 0 1 | 1 2 | 2 3 | 3 [...] | [...] 4 seconds later If we spend too much time to read the buffers (case of a lot of events to record in each buffers or when we have a lot of CPU buffers to read), in the next pass the CPU 0 buffer could contain a slice of several seconds of events. We'll read them all and notice we've reached the period to flush. In the above example we flush the first half of the CPU 0 buffer, then we read the CPU 1 buffer where we have events that were on the flush slice and then the reordering fails. It's simple to reproduce with: perf lock record perf bench sched messaging To solve this, we use a new solution that doesn't rely on an heuristical time slice period anymore but on a deterministic basis based on how perf record does its job. perf record saves the buffers through passes. A pass is a tour on every buffers from every CPUs. This is made in order: for each CPU we read the buffers of every counters. So the more buffers we visit, the later will be the timstamps of their events. When perf record finishes a pass it records a PERF_RECORD_FINISHED_ROUND pseudo event. We record the max timestamp t found in the pass n. Assuming these timestamps are monotonic across cpus, we know that if a buffer still has events with timestamps below t, they will be all available and then read in the pass n + 1. Hence when we start to read the pass n + 2, we can safely flush every events with timestamps below t. ============ PASS n ================= CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 1 | 2 2 | 3 - | 4 <--- max recorded ============ PASS n + 1 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 3 | 5 4 | 6 5 | 7 <---- max recorded Flush every events below timestamp 4 ============ PASS n + 2 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 6 | 8 7 | 9 - | 10 Flush every events below timestamp 7 etc... It also works on perf.data versions that don't have PERF_RECORD_FINISHED_ROUND pseudo events. The difference is that the events will be only flushed in the end of the perf.data processing. It will then consume more memory and scale less with large perf.data files. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Tom Zanussi <tzanussi@gmail.com> Cc: Masami Hiramatsu <mhiramat@redhat.com>
2010-05-03 21:14:33 +08:00
event_op sample,
mmap,
comm,
fork,
exit,
lost,
read,
throttle,
unthrottle;
event_synth_op attr,
perf: Provide a new deterministic events reordering algorithm The current events reordering algorithm is based on a heuristic that gets broken once we deal with a very fast flow of events. Indeed the time period based flushing is not suitable anymore in the following case, assuming we have a flush period of two seconds. CPU 0 | CPU 1 | cnt1 timestamps | cnt1 timestamps | 0 | 0 1 | 1 2 | 2 3 | 3 [...] | [...] 4 seconds later If we spend too much time to read the buffers (case of a lot of events to record in each buffers or when we have a lot of CPU buffers to read), in the next pass the CPU 0 buffer could contain a slice of several seconds of events. We'll read them all and notice we've reached the period to flush. In the above example we flush the first half of the CPU 0 buffer, then we read the CPU 1 buffer where we have events that were on the flush slice and then the reordering fails. It's simple to reproduce with: perf lock record perf bench sched messaging To solve this, we use a new solution that doesn't rely on an heuristical time slice period anymore but on a deterministic basis based on how perf record does its job. perf record saves the buffers through passes. A pass is a tour on every buffers from every CPUs. This is made in order: for each CPU we read the buffers of every counters. So the more buffers we visit, the later will be the timstamps of their events. When perf record finishes a pass it records a PERF_RECORD_FINISHED_ROUND pseudo event. We record the max timestamp t found in the pass n. Assuming these timestamps are monotonic across cpus, we know that if a buffer still has events with timestamps below t, they will be all available and then read in the pass n + 1. Hence when we start to read the pass n + 2, we can safely flush every events with timestamps below t. ============ PASS n ================= CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 1 | 2 2 | 3 - | 4 <--- max recorded ============ PASS n + 1 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 3 | 5 4 | 6 5 | 7 <---- max recorded Flush every events below timestamp 4 ============ PASS n + 2 ============== CPU 0 | CPU 1 | cnt1 timestamps | cnt2 timestamps 6 | 8 7 | 9 - | 10 Flush every events below timestamp 7 etc... It also works on perf.data versions that don't have PERF_RECORD_FINISHED_ROUND pseudo events. The difference is that the events will be only flushed in the end of the perf.data processing. It will then consume more memory and scale less with large perf.data files. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Tom Zanussi <tzanussi@gmail.com> Cc: Masami Hiramatsu <mhiramat@redhat.com>
2010-05-03 21:14:33 +08:00
event_type,
tracing_data,
build_id;
event_op2 finished_round;
bool ordered_samples;
perf session: Fallback to unordered processing if no sample_id_all If we are running the new perf on an old kernel without support for sample_id_all, we should fall back to the old unordered processing of events. If we didn't than we would *always* process events without timestamps out of order, whether or not we hit a reordering race. In other words, instead of there being a chance of not attributing samples correctly, we would guarantee that samples would not be attributed. While processing all events without timestamps before events with timestamps may seem like an intuitive solution, it falls down as PERF_RECORD_EXIT events would also be processed before any samples. Even with a workaround for that case, samples before/after an exec would not be attributed correctly. This patch allows commands to indicate whether they need to fall back to unordered processing, so that commands that do not care about timestamps on every event will not be affected. If we do fallback, this will print out a warning if report -D was invoked. This patch adds the test in perf_session__new so that we only need to test once per session. Commands that do not use an event_ops (such as record and top) can simply pass NULL in it's place. Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> LKML-Reference: <1291951882-sup-6069@au1.ibm.com> Signed-off-by: Ian Munsie <imunsie@au1.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-12-10 11:09:16 +08:00
bool ordering_requires_timestamps;
};
perf session: Fallback to unordered processing if no sample_id_all If we are running the new perf on an old kernel without support for sample_id_all, we should fall back to the old unordered processing of events. If we didn't than we would *always* process events without timestamps out of order, whether or not we hit a reordering race. In other words, instead of there being a chance of not attributing samples correctly, we would guarantee that samples would not be attributed. While processing all events without timestamps before events with timestamps may seem like an intuitive solution, it falls down as PERF_RECORD_EXIT events would also be processed before any samples. Even with a workaround for that case, samples before/after an exec would not be attributed correctly. This patch allows commands to indicate whether they need to fall back to unordered processing, so that commands that do not care about timestamps on every event will not be affected. If we do fallback, this will print out a warning if report -D was invoked. This patch adds the test in perf_session__new so that we only need to test once per session. Commands that do not use an event_ops (such as record and top) can simply pass NULL in it's place. Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> LKML-Reference: <1291951882-sup-6069@au1.ibm.com> Signed-off-by: Ian Munsie <imunsie@au1.ibm.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-12-10 11:09:16 +08:00
struct perf_session *perf_session__new(const char *filename, int mode,
bool force, bool repipe,
struct perf_event_ops *ops);
void perf_session__delete(struct perf_session *self);
perf tools: Cross platform perf.data analysis support There are still some problems related to loading vmlinux files, but those are unrelated to the feature implemented in this patch, so will get fixed in the next patches, but here are some results: 1. collect perf.data file on a Fedora 12 machine, x86_64, 64-bit userland 2. transfer it to a Debian Testing machine, PARISC64, 32-bit userland acme@parisc:~/git/linux-2.6-tip$ perf buildid-list | head -5 74f9930ee94475b6b3238caf3725a50d59cb994b [kernel.kallsyms] 55fdd56670453ea66c011158c4b9d30179c1d049 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/ipt_MASQUERADE.ko 41adff63c730890480980d5d8ba513f1c216a858 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/iptable_nat.ko 90a33def1077bb8e97b8a78546dc96c2de62df46 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/nf_nat.ko 984c7bea90ce1376d5c8e7ef43a781801286e62d /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/tun.ko acme@parisc:~/git/linux-2.6-tip$ perf buildid-list | tail -5 22492f3753c6a67de5c7ccbd6b863390c92c0723 /usr/lib64/libXt.so.6.0.0 353802bb7e1b895ba43507cc678f951e778e4c6f /usr/lib64/libMagickCore.so.2.0.0 d10c2897558595efe7be8b0584cf7e6398bc776c /usr/lib64/libfprint.so.0.0.0 a83ecfb519a788774a84d5ddde633c9ba56c03ab /home/acme/bin/perf d3ca765a8ecf257d263801d7ad8c49c189082317 /usr/lib64/libdwarf.so.0.0 acme@parisc:~/git/linux-2.6-tip$ acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm The file [kernel.kallsyms] cannot be used, trying to use /proc/kallsyms... ^^^^ The problem related to vmlinux handling, it shouldn't be trying this ^^^^ rather alien /proc/kallsyms at all... /lib64/libpthread-2.10.2.so with build id 5c68f7afeb33309c78037e374b0deee84dd441f6 not found, continuing without symbols /lib64/libc-2.10.2.so with build id eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 not found, continuing without symbols /home/acme/bin/perf with build id a83ecfb519a788774a84d5ddde633c9ba56c03ab not found, continuing without symbols /usr/sbin/openvpn with build id f2037a091ef36b591187a858d75e203690ea9409 not found, continuing without symbols Failed to open /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/e1000e/e1000e.ko, continuing without symbols Failed to open /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/wireless/iwlwifi/iwlcore.ko, continuing without symbols <SNIP more complaints about not finding the right build-ids, those will have to wait for 'perf archive' or plain copying what was collected by 'perf record' on the x86_64, source machine, see further below for an example of this > # Samples: 293085637 # # Overhead Command # ........ ............... # 61.70% find 23.50% perf 5.86% swapper 3.12% sshd 2.39% init 0.87% bash 0.86% sleep 0.59% dbus-daemon 0.25% hald 0.24% NetworkManager 0.19% hald-addon-rfki 0.15% openvpn 0.07% phy0 0.07% events/0 0.05% iwl3945 0.05% events/1 0.03% kondemand/0 acme@parisc:~/git/linux-2.6-tip$ Which matches what we get when running the same command for the same perf.data file on the F12, x86_64, source machine: [root@doppio linux-2.6-tip]# perf report --sort comm # Samples: 293085637 # # Overhead Command # ........ ............... # 61.70% find 23.50% perf 5.86% swapper 3.12% sshd 2.39% init 0.87% bash 0.86% sleep 0.59% dbus-daemon 0.25% hald 0.24% NetworkManager 0.19% hald-addon-rfki 0.15% openvpn 0.07% phy0 0.07% events/0 0.05% iwl3945 0.05% events/1 0.03% kondemand/0 [root@doppio linux-2.6-tip]# The other modes work as well, modulo the problem with vmlinux: acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm,dso 2> /dev/null | head -15 # Samples: 293085637 # # Overhead Command Shared Object # ........ ............... ................................. # 35.11% find ffffffff81002b5a 18.25% perf ffffffff8102235f 16.17% find libc-2.10.2.so 9.07% find find 5.80% swapper ffffffff8102235f 3.95% perf libc-2.10.2.so 2.33% init ffffffff810091b9 1.65% sshd libcrypto.so.0.9.8k 1.35% find [e1000e] 0.68% sleep libc-2.10.2.so acme@parisc:~/git/linux-2.6-tip$ And the lack of the right buildids: acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm,dso,symbol 2> /dev/null | head -15 # Samples: 293085637 # # Overhead Command Shared Object Symbol # ........ ............... ................................. ...... # 35.11% find ffffffff81002b5a [k] 0xffffffff81002b5a 18.25% perf ffffffff8102235f [k] 0xffffffff8102235f 16.17% find libc-2.10.2.so [.] 0x00000000045782 9.07% find find [.] 0x0000000000fb0e 5.80% swapper ffffffff8102235f [k] 0xffffffff8102235f 3.95% perf libc-2.10.2.so [.] 0x0000000007f398 2.33% init ffffffff810091b9 [k] 0xffffffff810091b9 1.65% sshd libcrypto.so.0.9.8k [.] 0x00000000105440 1.35% find [e1000e] [k] 0x00000000010948 0.68% sleep libc-2.10.2.so [.] 0x0000000011ad5b acme@parisc:~/git/linux-2.6-tip$ But if we: acme@parisc:~/git/linux-2.6-tip$ ls ~/.debug ls: cannot access /home/acme/.debug: No such file or directory acme@parisc:~/git/linux-2.6-tip$ mkdir -p ~/.debug/lib64/libc-2.10.2.so/ acme@parisc:~/git/linux-2.6-tip$ scp doppio:.debug/lib64/libc-2.10.2.so/* ~/.debug/lib64/libc-2.10.2.so/ acme@doppio's password: eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 100% 1783KB 714.7KB/s 00:02 acme@parisc:~/git/linux-2.6-tip$ mkdir -p ~/.debug/.build-id/eb acme@parisc:~/git/linux-2.6-tip$ ln -s ../../lib64/libc-2.10.2.so/eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 ~/.debug/.build-id/eb/4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 acme@parisc:~/git/linux-2.6-tip$ perf report --dsos libc-2.10.2.so 2> /dev/null # dso: libc-2.10.2.so # Samples: 64281170 # # Overhead Command Symbol # ........ ............... ...... # 14.98% perf [.] __GI_strcmp 12.30% find [.] __GI_memmove 9.25% find [.] _int_malloc 7.60% find [.] _IO_vfprintf_internal 6.10% find [.] _IO_new_file_xsputn 6.02% find [.] __GI_close 3.08% find [.] _IO_file_overflow_internal 3.08% find [.] malloc_consolidate 3.08% find [.] _int_free 3.08% find [.] __strchrnul 3.08% find [.] __getdents64 3.08% find [.] __write_nocancel 3.08% sleep [.] __GI__dl_addr 3.08% sshd [.] __libc_select 3.08% find [.] _IO_new_file_write 3.07% find [.] _IO_new_do_write 3.06% find [.] __GI___errno_location 3.05% find [.] __GI___libc_malloc 3.04% perf [.] __GI_memcpy 1.71% find [.] __fprintf_chk 1.29% bash [.] __gconv_transform_utf8_internal 0.79% dbus-daemon [.] __GI_strlen # # (For a higher level overview, try: perf report --sort comm,dso) # acme@parisc:~/git/linux-2.6-tip$ Which matches what we get on the source, F12, x86_64 machine: [root@doppio linux-2.6-tip]# perf report --dsos libc-2.10.2.so # dso: libc-2.10.2.so # Samples: 64281170 # # Overhead Command Symbol # ........ ............... ...... # 14.98% perf [.] __GI_strcmp 12.30% find [.] __GI_memmove 9.25% find [.] _int_malloc 7.60% find [.] _IO_vfprintf_internal 6.10% find [.] _IO_new_file_xsputn 6.02% find [.] __GI_close 3.08% find [.] _IO_file_overflow_internal 3.08% find [.] malloc_consolidate 3.08% find [.] _int_free 3.08% find [.] __strchrnul 3.08% find [.] __getdents64 3.08% find [.] __write_nocancel 3.08% sleep [.] __GI__dl_addr 3.08% sshd [.] __libc_select 3.08% find [.] _IO_new_file_write 3.07% find [.] _IO_new_do_write 3.06% find [.] __GI___errno_location 3.05% find [.] __GI___libc_malloc 3.04% perf [.] __GI_memcpy 1.71% find [.] __fprintf_chk 1.29% bash [.] __gconv_transform_utf8_internal 0.79% dbus-daemon [.] __GI_strlen # # (For a higher level overview, try: perf report --sort comm,dso) # [root@doppio linux-2.6-tip]# So I think this is really, really nice in that it demonstrates the portability of perf.data files and the use of build-ids accross such aliens worlds :-) There are some things to fix tho, like the bitmap on the header, but things are looking good. 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: <1263478990-8200-2-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-01-14 22:23:10 +08:00
void perf_event_header__bswap(struct perf_event_header *self);
int __perf_session__process_events(struct perf_session *self,
u64 data_offset, u64 data_size, u64 size,
struct perf_event_ops *ops);
int perf_session__process_events(struct perf_session *self,
struct perf_event_ops *event_ops);
perf callchain: Feed callchains into a cursor The callchains are fed with an array of a fixed size. As a result we iterate over each callchains three times: - 1st to resolve symbols - 2nd to filter out context boundaries - 3rd for the insertion into the tree This also involves some pairs of memory allocation/deallocation everytime we insert a callchain, for the filtered out array of addresses and for the array of symbols that comes along. Instead, feed the callchains through a linked list with persistent allocations. It brings several pros like: - Merge the 1st and 2nd iterations in one. That was possible before but in a way that would involve allocating an array slightly taller than necessary because we don't know in advance the number of context boundaries to filter out. - Much lesser allocations/deallocations. The linked list keeps persistent empty entries for the next usages and is extendable at will. - Makes it easier for multiple sources of callchains to feed a stacktrace together. This is deemed to pave the way for cfi based callchains wherein traditional frame pointer based kernel stacktraces will precede cfi based user ones, producing an overall callchain which size is hardly predictable. This requirement makes the static array obsolete and makes a linked list based iterator a much more flexible fit. Basic testing on a big perf file containing callchains (~ 176 MB) has shown a throughput gain of about 11% with perf report. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1294977121-5700-2-git-send-email-fweisbec@gmail.com> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-01-14 11:51:58 +08:00
int perf_session__resolve_callchain(struct perf_session *self,
struct thread *thread,
struct ip_callchain *chain,
struct symbol **parent);
bool perf_session__has_traces(struct perf_session *self, const char *msg);
int perf_session__set_kallsyms_ref_reloc_sym(struct map **maps,
perf tools: Handle relocatable kernels DSOs don't have this problem because the kernel emits a PERF_MMAP for each new executable mapping it performs on monitored threads. To fix the kernel case we simulate the same behaviour, by having 'perf record' to synthesize a PERF_MMAP for the kernel, encoded like this: [root@doppio ~]# perf record -a -f sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.344 MB perf.data (~15038 samples) ] [root@doppio ~]# perf report -D | head -10 0xd0 [0x40]: event: 1 . . ... raw event: size 64 bytes . 0000: 01 00 00 00 00 00 40 00 00 00 00 00 00 00 00 00 ......@........ . 0010: 00 00 00 81 ff ff ff ff 00 00 00 00 00 00 00 00 ............... . 0020: 00 00 00 00 00 00 00 00 5b 6b 65 72 6e 65 6c 2e ........ [kernel . 0030: 6b 61 6c 6c 73 79 6d 73 2e 5f 74 65 78 74 5d 00 kallsyms._text] . 0xd0 [0x40]: PERF_RECORD_MMAP 0/0: [0xffffffff81000000((nil)) @ (nil)]: [kernel.kallsyms._text] I.e. we identify such event as having: .pid = 0 .filename = [kernel.kallsyms.REFNAME] .start = REFNAME addr in /proc/kallsyms at 'perf record' time and use now a hardcoded value of '.text' for REFNAME. Then, later, in 'perf report', if there are any kernel hits and thus we need to resolve kernel symbols, we search for REFNAME and if its address changed, relocation happened and we thus must change the kernel mapping routines to one that uses .pgoff as the relocation to apply. This way we use the same mechanism used for the other DSOs and don't have to do a two pass in all the kernel symbols. Reported-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> LKML-Reference: <1262717431-1246-1-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-01-06 02:50:31 +08:00
const char *symbol_name,
u64 addr);
perf tools: Cross platform perf.data analysis support There are still some problems related to loading vmlinux files, but those are unrelated to the feature implemented in this patch, so will get fixed in the next patches, but here are some results: 1. collect perf.data file on a Fedora 12 machine, x86_64, 64-bit userland 2. transfer it to a Debian Testing machine, PARISC64, 32-bit userland acme@parisc:~/git/linux-2.6-tip$ perf buildid-list | head -5 74f9930ee94475b6b3238caf3725a50d59cb994b [kernel.kallsyms] 55fdd56670453ea66c011158c4b9d30179c1d049 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/ipt_MASQUERADE.ko 41adff63c730890480980d5d8ba513f1c216a858 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/iptable_nat.ko 90a33def1077bb8e97b8a78546dc96c2de62df46 /lib/modules/2.6.33-rc4-tip+/kernel/net/ipv4/netfilter/nf_nat.ko 984c7bea90ce1376d5c8e7ef43a781801286e62d /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/tun.ko acme@parisc:~/git/linux-2.6-tip$ perf buildid-list | tail -5 22492f3753c6a67de5c7ccbd6b863390c92c0723 /usr/lib64/libXt.so.6.0.0 353802bb7e1b895ba43507cc678f951e778e4c6f /usr/lib64/libMagickCore.so.2.0.0 d10c2897558595efe7be8b0584cf7e6398bc776c /usr/lib64/libfprint.so.0.0.0 a83ecfb519a788774a84d5ddde633c9ba56c03ab /home/acme/bin/perf d3ca765a8ecf257d263801d7ad8c49c189082317 /usr/lib64/libdwarf.so.0.0 acme@parisc:~/git/linux-2.6-tip$ acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm The file [kernel.kallsyms] cannot be used, trying to use /proc/kallsyms... ^^^^ The problem related to vmlinux handling, it shouldn't be trying this ^^^^ rather alien /proc/kallsyms at all... /lib64/libpthread-2.10.2.so with build id 5c68f7afeb33309c78037e374b0deee84dd441f6 not found, continuing without symbols /lib64/libc-2.10.2.so with build id eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 not found, continuing without symbols /home/acme/bin/perf with build id a83ecfb519a788774a84d5ddde633c9ba56c03ab not found, continuing without symbols /usr/sbin/openvpn with build id f2037a091ef36b591187a858d75e203690ea9409 not found, continuing without symbols Failed to open /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/e1000e/e1000e.ko, continuing without symbols Failed to open /lib/modules/2.6.33-rc4-tip+/kernel/drivers/net/wireless/iwlwifi/iwlcore.ko, continuing without symbols <SNIP more complaints about not finding the right build-ids, those will have to wait for 'perf archive' or plain copying what was collected by 'perf record' on the x86_64, source machine, see further below for an example of this > # Samples: 293085637 # # Overhead Command # ........ ............... # 61.70% find 23.50% perf 5.86% swapper 3.12% sshd 2.39% init 0.87% bash 0.86% sleep 0.59% dbus-daemon 0.25% hald 0.24% NetworkManager 0.19% hald-addon-rfki 0.15% openvpn 0.07% phy0 0.07% events/0 0.05% iwl3945 0.05% events/1 0.03% kondemand/0 acme@parisc:~/git/linux-2.6-tip$ Which matches what we get when running the same command for the same perf.data file on the F12, x86_64, source machine: [root@doppio linux-2.6-tip]# perf report --sort comm # Samples: 293085637 # # Overhead Command # ........ ............... # 61.70% find 23.50% perf 5.86% swapper 3.12% sshd 2.39% init 0.87% bash 0.86% sleep 0.59% dbus-daemon 0.25% hald 0.24% NetworkManager 0.19% hald-addon-rfki 0.15% openvpn 0.07% phy0 0.07% events/0 0.05% iwl3945 0.05% events/1 0.03% kondemand/0 [root@doppio linux-2.6-tip]# The other modes work as well, modulo the problem with vmlinux: acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm,dso 2> /dev/null | head -15 # Samples: 293085637 # # Overhead Command Shared Object # ........ ............... ................................. # 35.11% find ffffffff81002b5a 18.25% perf ffffffff8102235f 16.17% find libc-2.10.2.so 9.07% find find 5.80% swapper ffffffff8102235f 3.95% perf libc-2.10.2.so 2.33% init ffffffff810091b9 1.65% sshd libcrypto.so.0.9.8k 1.35% find [e1000e] 0.68% sleep libc-2.10.2.so acme@parisc:~/git/linux-2.6-tip$ And the lack of the right buildids: acme@parisc:~/git/linux-2.6-tip$ perf report --sort comm,dso,symbol 2> /dev/null | head -15 # Samples: 293085637 # # Overhead Command Shared Object Symbol # ........ ............... ................................. ...... # 35.11% find ffffffff81002b5a [k] 0xffffffff81002b5a 18.25% perf ffffffff8102235f [k] 0xffffffff8102235f 16.17% find libc-2.10.2.so [.] 0x00000000045782 9.07% find find [.] 0x0000000000fb0e 5.80% swapper ffffffff8102235f [k] 0xffffffff8102235f 3.95% perf libc-2.10.2.so [.] 0x0000000007f398 2.33% init ffffffff810091b9 [k] 0xffffffff810091b9 1.65% sshd libcrypto.so.0.9.8k [.] 0x00000000105440 1.35% find [e1000e] [k] 0x00000000010948 0.68% sleep libc-2.10.2.so [.] 0x0000000011ad5b acme@parisc:~/git/linux-2.6-tip$ But if we: acme@parisc:~/git/linux-2.6-tip$ ls ~/.debug ls: cannot access /home/acme/.debug: No such file or directory acme@parisc:~/git/linux-2.6-tip$ mkdir -p ~/.debug/lib64/libc-2.10.2.so/ acme@parisc:~/git/linux-2.6-tip$ scp doppio:.debug/lib64/libc-2.10.2.so/* ~/.debug/lib64/libc-2.10.2.so/ acme@doppio's password: eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 100% 1783KB 714.7KB/s 00:02 acme@parisc:~/git/linux-2.6-tip$ mkdir -p ~/.debug/.build-id/eb acme@parisc:~/git/linux-2.6-tip$ ln -s ../../lib64/libc-2.10.2.so/eb4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 ~/.debug/.build-id/eb/4ec8fa8b2a5eb18cad173c92f27ed8887ed1c1 acme@parisc:~/git/linux-2.6-tip$ perf report --dsos libc-2.10.2.so 2> /dev/null # dso: libc-2.10.2.so # Samples: 64281170 # # Overhead Command Symbol # ........ ............... ...... # 14.98% perf [.] __GI_strcmp 12.30% find [.] __GI_memmove 9.25% find [.] _int_malloc 7.60% find [.] _IO_vfprintf_internal 6.10% find [.] _IO_new_file_xsputn 6.02% find [.] __GI_close 3.08% find [.] _IO_file_overflow_internal 3.08% find [.] malloc_consolidate 3.08% find [.] _int_free 3.08% find [.] __strchrnul 3.08% find [.] __getdents64 3.08% find [.] __write_nocancel 3.08% sleep [.] __GI__dl_addr 3.08% sshd [.] __libc_select 3.08% find [.] _IO_new_file_write 3.07% find [.] _IO_new_do_write 3.06% find [.] __GI___errno_location 3.05% find [.] __GI___libc_malloc 3.04% perf [.] __GI_memcpy 1.71% find [.] __fprintf_chk 1.29% bash [.] __gconv_transform_utf8_internal 0.79% dbus-daemon [.] __GI_strlen # # (For a higher level overview, try: perf report --sort comm,dso) # acme@parisc:~/git/linux-2.6-tip$ Which matches what we get on the source, F12, x86_64 machine: [root@doppio linux-2.6-tip]# perf report --dsos libc-2.10.2.so # dso: libc-2.10.2.so # Samples: 64281170 # # Overhead Command Symbol # ........ ............... ...... # 14.98% perf [.] __GI_strcmp 12.30% find [.] __GI_memmove 9.25% find [.] _int_malloc 7.60% find [.] _IO_vfprintf_internal 6.10% find [.] _IO_new_file_xsputn 6.02% find [.] __GI_close 3.08% find [.] _IO_file_overflow_internal 3.08% find [.] malloc_consolidate 3.08% find [.] _int_free 3.08% find [.] __strchrnul 3.08% find [.] __getdents64 3.08% find [.] __write_nocancel 3.08% sleep [.] __GI__dl_addr 3.08% sshd [.] __libc_select 3.08% find [.] _IO_new_file_write 3.07% find [.] _IO_new_do_write 3.06% find [.] __GI___errno_location 3.05% find [.] __GI___libc_malloc 3.04% perf [.] __GI_memcpy 1.71% find [.] __fprintf_chk 1.29% bash [.] __gconv_transform_utf8_internal 0.79% dbus-daemon [.] __GI_strlen # # (For a higher level overview, try: perf report --sort comm,dso) # [root@doppio linux-2.6-tip]# So I think this is really, really nice in that it demonstrates the portability of perf.data files and the use of build-ids accross such aliens worlds :-) There are some things to fix tho, like the bitmap on the header, but things are looking good. 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: <1263478990-8200-2-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-01-14 22:23:10 +08:00
void mem_bswap_64(void *src, int byte_size);
int perf_session__create_kernel_maps(struct perf_session *self);
perf report: Implement initial UI using newt Newt has widespread availability and provides a rather simple API as can be seen by the size of this patch. The work needed to support it will benefit other frontends too. In this initial patch it just checks if the output is a tty, if not it falls back to the previous behaviour, also if newt-devel/libnewt-dev is not installed the previous behaviour is maintaned. Pressing enter on a symbol will annotate it, ESC in the annotation window will return to the report symbol list. More work will be done to remove the special casing in color_fprintf, stop using fmemopen/FILE in the printing of hist_entries, etc. Also the annotation doesn't need to be done via spawning "perf annotate" and then browsing its output, we can do better by calling directly the builtin-annotate.c functions, that would then be moved to tools/perf/util/annotate.c and shared with perf top, etc But lets go by baby steps, this patch already improves perf usability by allowing to quickly do annotations on symbols from the report screen and provides a first experimentation with libnewt/TUI integration of tools. Tested on RHEL5 and Fedora12 X86_64 and on Debian PARISC64 to browse a perf.data file collected on a Fedora12 x86_64 box. Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Avi Kivity <avi@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: <1268349164-5822-5-git-send-email-acme@infradead.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-03-12 07:12:44 +08:00
void perf_session__update_sample_type(struct perf_session *self);
perf tools: Ask for ID PERF_SAMPLE_ info on all PERF_RECORD_ events So that we can use -T == --timestamp, asking for PERF_SAMPLE_TIME: $ perf record -aT $ perf report -D | grep PERF_RECORD_ <SNIP> 3 5951915425 0x47530 [0x58]: PERF_RECORD_SAMPLE(IP, 1): 16811/16811: 0xffffffff8138c1a2 period: 215979 cpu:3 3 5952026879 0x47588 [0x90]: PERF_RECORD_SAMPLE(IP, 1): 16811/16811: 0xffffffff810cb480 period: 215979 cpu:3 3 5952059959 0x47618 [0x38]: PERF_RECORD_FORK(6853:6853):(16811:16811) 3 5952138878 0x47650 [0x78]: PERF_RECORD_SAMPLE(IP, 1): 16811/16811: 0xffffffff811bac35 period: 431478 cpu:3 3 5952375068 0x476c8 [0x30]: PERF_RECORD_COMM: find:6853 3 5952395923 0x476f8 [0x50]: PERF_RECORD_MMAP 6853/6853: [0x400000(0x25000) @ 0]: /usr/bin/find 3 5952413756 0x47748 [0xa0]: PERF_RECORD_SAMPLE(IP, 1): 6853/6853: 0xffffffff810d080f period: 859332 cpu:3 3 5952419837 0x477e8 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f44600000(0x21d000) @ 0]: /lib64/ld-2.5.so 3 5952437929 0x47840 [0x48]: PERF_RECORD_MMAP 6853/6853: [0x7fff7e1c9000(0x1000) @ 0x7fff7e1c9000]: [vdso] 3 5952570127 0x47888 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f46200000(0x218000) @ 0]: /lib64/libselinux.so.1 3 5952623637 0x478e0 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f44a00000(0x356000) @ 0]: /lib64/libc-2.5.so 3 5952675720 0x47938 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f44e00000(0x204000) @ 0]: /lib64/libdl-2.5.so 3 5952710080 0x47990 [0x58]: PERF_RECORD_MMAP 6853/6853: [0x3f45a00000(0x246000) @ 0]: /lib64/libsepol.so.1 3 5952847802 0x479e8 [0x58]: PERF_RECORD_SAMPLE(IP, 1): 6853/6853: 0xffffffff813897f0 period: 1142536 cpu:3 <SNIP> First column is the cpu and the second the timestamp. That way we can investigate problems in the event stream. If the new perf binary is run on an older kernel, it will disable this feature automatically. Tested-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ian Munsie <imunsie@au1.ibm.com> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: Ian Munsie <imunsie@au1.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> Cc: Stephane Eranian <eranian@google.com> LKML-Reference: <1291318772-30880-5-git-send-email-acme@infradead.org> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2010-12-02 20:25:28 +08:00
void perf_session__set_sample_id_all(struct perf_session *session, bool value);
void perf_session__set_sample_type(struct perf_session *session, u64 type);
void perf_session__remove_thread(struct perf_session *self, struct thread *th);
static inline
struct machine *perf_session__find_host_machine(struct perf_session *self)
{
return &self->host_machine;
}
static inline
struct machine *perf_session__find_machine(struct perf_session *self, pid_t pid)
{
if (pid == HOST_KERNEL_ID)
return &self->host_machine;
return machines__find(&self->machines, pid);
}
static inline
struct machine *perf_session__findnew_machine(struct perf_session *self, pid_t pid)
{
if (pid == HOST_KERNEL_ID)
return &self->host_machine;
return machines__findnew(&self->machines, pid);
}
static inline
void perf_session__process_machines(struct perf_session *self,
machine__process_t process)
{
process(&self->host_machine, self);
return machines__process(&self->machines, process, self);
}
size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp);
size_t perf_session__fprintf_dsos_buildid(struct perf_session *self,
FILE *fp, bool with_hits);
static inline
size_t perf_session__fprintf_nr_events(struct perf_session *self, FILE *fp)
{
return hists__fprintf_nr_events(&self->hists, fp);
}
static inline int perf_session__parse_sample(struct perf_session *session,
const union perf_event *event,
struct perf_sample *sample)
{
return perf_event__parse_sample(event, session->sample_type,
session->sample_id_all, sample);
}
#endif /* __PERF_SESSION_H */