[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <1415653280-1374-1-git-send-email-kan.liang@intel.com>
Date: Mon, 10 Nov 2014 16:01:17 -0500
From: kan.liang@...el.com
To: acme@...nel.org, jolsa@...hat.com, namhyung@...nel.org
Cc: linux-kernel@...r.kernel.org, ak@...ux.intel.com,
Kan Liang <kan.liang@...el.com>
Subject: [PATCH V2 0/3] perf tool: perf diff sort changes
From: Kan Liang <kan.liang@...el.com>
Current perf diff has some issues. E.g. the default sort key is
symbol, but it doesn't work well. It sorts as address. Furthermore,
the old perf diff can only work on perf.data from the same binary.
This patch set fixes the symbol issue and extends it to compare the
perf.data from different binaries and different kernels. It's useful
for debugging the regression issue.
Current perf diff can do address compare, the patch set also keep
this feature. A new sort key "addr" is introduced. It can let the user
sort differential profile by the address. It should be useful for
debugging the scaling issue, if the user think function level diff is
too high granularity.
Here is an example.
v1_1_6perf.data is perf record result of version one tchain on 1.6 kernel.
v2_1_8perf.data is perf record result of version two tchain on 1.8 kernel.
They are from different binaries.
v1_1_8_1perf.data and v1_1_8_2perf.data are perf record result of version
one tchain on 1.8 kernel. I run perf record twice.
They are from same binary.
Old perf diff with default sort key "dso,symbol" for different binary
./perf diff -s dso,symbol v1_1_6perf.data v2_1_8perf.data
Event 'cycles'
Baseline Delta Shared Object Symbol
........ ....... ................. .........................
0.01% [kernel.kallsyms] [k] native_write_msr_safe
0.01% [kernel.kallsyms] [k] notifier_call_chain
0.01% [kernel.kallsyms] [k] perf_event_task_tick
0.01% [kernel.kallsyms] [k] run_posix_cpu_timers
0.01% [kernel.kallsyms] [k] run_timer_softirq
0.01% [kernel.kallsyms] [k] trigger_load_balance
0.01% [kernel.kallsyms] [k] update_vsyscall
[kernel.vmlinux] [k] __run_hrtimer
[kernel.vmlinux] [k] apic_timer_interrupt
[kernel.vmlinux] [k] enqueue_task
[kernel.vmlinux] [k] hrtimer_interrupt
[kernel.vmlinux] [k] native_write_msr_safe
[kernel.vmlinux] [k] trigger_load_balance
[kernel.vmlinux] [k] update_wall_time
0.05% [unknown] [.] 0x0000000000400540
0.04% [unknown] [.] 0x0000000000400541
0.03% [unknown] [.] 0x000000000040054b
0.04% [unknown] [.] 0x0000000000400552
33.55% [unknown] [.] 0x0000000000400554
1.22% [unknown] [.] 0x000000000040055a
8.00% [unknown] [.] 0x000000000040055e
0.02% [unknown] [.] 0x0000000000400562
8.41% [unknown] [.] 0x0000000000400564
48.13% [unknown] [.] 0x000000000040056b
0.16% [unknown] [.] 0x0000000000400570
0.17% [unknown] [.] 0x0000000000400571
+0.45% [unknown] [.] 0x0000000000400580
+0.29% [unknown] [.] 0x0000000000400581
0.01% [unknown] [.] 0x0000000000400583
0.01% [unknown] [.] 0x0000000000400588
+0.22% [unknown] [.] 0x000000000040058b
0.01% +13.35% [unknown] [.] 0x000000000040058d
0.06% [unknown] [.] 0x0000000000400591
+0.78% [unknown] [.] 0x0000000000400593
0.04% [unknown] [.] 0x0000000000400595
0.01% +6.18% [unknown] [.] 0x0000000000400597
+1.47% [unknown] [.] 0x000000000040059b
+6.46% [unknown] [.] 0x000000000040059d
+1.28% [unknown] [.] 0x00000000004005a1
+66.38% [unknown] [.] 0x00000000004005a5
+1.34% [unknown] [.] 0x00000000004005a7
+1.15% [unknown] [.] 0x00000000004005a8
+0.05% [unknown] [.] 0x00000000004005ba
+0.03% [unknown] [.] 0x00000000004005bf
+0.02% [unknown] [.] 0x00000000004005c4
+0.05% [unknown] [.] 0x00000000004005c9
+0.03% [unknown] [.] 0x00000000004005ce
+0.27% [unknown] [.] 0x00000000004005d2
+0.15% [unknown] [.] 0x00000000004005d6
[unknown] [.] 0x00000000004005d8
[unknown] [.] 0x00000000004005d9
New perf diff with default sort key "dso,symbol" for different binary
./perf diff -s dso,symbol v1_1_6perf.data v2_1_8perf.data
Event 'cycles'
Baseline Delta Shared Object Symbol
........ ....... ................. .........................
0.01% [kernel.kallsyms] [k] native_write_msr_safe
0.01% [kernel.kallsyms] [k] notifier_call_chain
0.01% [kernel.kallsyms] [k] perf_event_task_tick
0.01% [kernel.kallsyms] [k] run_posix_cpu_timers
0.01% [kernel.kallsyms] [k] run_timer_softirq
0.01% [kernel.kallsyms] [k] trigger_load_balance
0.01% [kernel.kallsyms] [k] update_vsyscall
[kernel.vmlinux] [k] __run_hrtimer
[kernel.vmlinux] [k] apic_timer_interrupt
[kernel.vmlinux] [k] enqueue_task
[kernel.vmlinux] [k] hrtimer_interrupt
[kernel.vmlinux] [k] native_write_msr_safe
[kernel.vmlinux] [k] trigger_load_balance
[kernel.vmlinux] [k] update_wall_time
0.14% +0.47% tchain [.] f2
99.82% -0.47% tchain [.] f3
New perf diff with sort key "symbol,addr" ffor same binary
perf diff -s symbol,addr v1_1_8_1perf.data v1_1_8_2perf.data
Event 'cycles'
Baseline Delta Symbol Addr
........ ....... ............................... ...............................
0.00% [k] __update_cpu_load [k] 0xffffffff810b9b8c
[k] _raw_spin_lock_irqsave [k] 0xffffffff81759364
[k] _raw_spin_unlock_irqrestore [k] 0xffffffff817594ba
0.00% [k] apic_timer_interrupt [k] 0xffffffff8175a9f0
[k] apic_timer_interrupt [k] 0xffffffff8175aa58
[k] check_preempt_curr [k] 0xffffffff810b461c
0.00% [.] f1 [.] 0x00000000004005fa
0.03% +0.03% [.] f2 [.] 0x00000000004005ba
0.03% +0.01% [.] f2 [.] 0x00000000004005bf
0.05% [.] f2 [.] 0x00000000004005c4
0.04% -0.02% [.] f2 [.] 0x00000000004005c9
0.03% [.] f2 [.] 0x00000000004005ce
0.17% +0.11% [.] f2 [.] 0x00000000004005d2
0.15% [.] f2 [.] 0x00000000004005d6
[.] f2 [.] 0x00000000004005d9
0.37% +0.02% [.] f3 [.] 0x0000000000400580
0.18% +0.03% [.] f3 [.] 0x0000000000400581
0.01% [.] f3 [.] 0x0000000000400584
0.18% +0.04% [.] f3 [.] 0x000000000040058b
12.31% +0.11% [.] f3 [.] 0x000000000040058d
0.03% [.] f3 [.] 0x0000000000400590
0.80% -0.12% [.] f3 [.] 0x0000000000400593
6.66% +0.09% [.] f3 [.] 0x0000000000400597
1.81% -0.36% [.] f3 [.] 0x000000000040059b
6.35% +0.24% [.] f3 [.] 0x000000000040059d
1.42% -0.22% [.] f3 [.] 0x00000000004005a1
66.83% +0.22% [.] f3 [.] 0x00000000004005a5
1.29% -0.12% [.] f3 [.] 0x00000000004005a7
1.22% -0.04% [.] f3 [.] 0x00000000004005a8
0.00% [k] load_balance [k] 0xffffffff810be7f6
0.00% [k] native_apic_mem_write [k] 0xffffffff81077920
0.00% [k] native_write_msr_safe [k] 0xffffffff8107d7ba
0.00% [k] native_write_msr_safe [k] 0xffffffff8107d7bd
0.00% [k] rb_erase [k] 0xffffffff813346b1
[k] resched_curr [k] 0xffffffff810b4095
[k] restore_args [k] 0xffffffff8175a744
0.00% [k] run_timer_softirq [k] 0xffffffff810e306f
[k] select_task_rq_fair [k] 0xffffffff810bbab9
0.00% [k] select_task_rq_fair [k] 0xffffffff810bbc89
[k] task_tick_fair [k] 0xffffffff810bd286
0.00% [k] task_tick_fair [k] 0xffffffff810bd40e
[k] task_waking_fair [k] 0xffffffff810baf97
0.00% [k] try_to_wake_up [k] 0xffffffff810b75e8
[k] update_cfs_rq_blocked_load [k] 0xffffffff810bb129
0.00% [k] update_cpu_load_active [k] 0xffffffff810ba08b
[k] update_group_capacity [k] 0xffffffff810bdde0
[k] update_wall_time [k] 0xffffffff810eac46
Changes from V1:
- mmap2 as default of tool's definition
- Using se_collapse to match symbol name.
- Sort key "addr" is introduced to compare address from same binary.
Kan Liang (3):
perf tool: Fix perf diff symble sort issue
perf tool:perf diff support for different binaries
perf tool: Add sort key addr for perf diff
tools/perf/Documentation/perf-diff.txt | 7 +++++--
tools/perf/builtin-diff.c | 3 ++-
tools/perf/util/hist.c | 5 +++--
tools/perf/util/hist.h | 1 +
tools/perf/util/sort.c | 38 ++++++++++++++++++++++++++++++++++
tools/perf/util/sort.h | 2 ++
6 files changed, 51 insertions(+), 5 deletions(-)
--
1.8.3.2
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists