ACPI: APEI: Fix integer overflow in ghes_estatus_pool_init()
[sfrench/cifs-2.6.git] / Documentation / tools / rtla / rtla-timerlat-top.rst
1 ====================
2 rtla-timerlat-top
3 ====================
4 -------------------------------------------
5 Measures the operating system timer latency
6 -------------------------------------------
7
8 :Manual section: 1
9
10 SYNOPSIS
11 ========
12 **rtla timerlat top** [*OPTIONS*] ...
13
14 DESCRIPTION
15 ===========
16
17 .. include:: common_timerlat_description.rst
18
19 The **rtla timerlat top** displays a summary of the periodic output
20 from the *timerlat* tracer. It also provides information for each
21 operating system noise via the **osnoise:** tracepoints that can be
22 seem with the option **-T**.
23
24 OPTIONS
25 =======
26
27 .. include:: common_timerlat_options.rst
28
29 .. include:: common_top_options.rst
30
31 .. include:: common_options.rst
32
33 EXAMPLE
34 =======
35
36 In the example below, the *timerlat* tracer is set to capture the stack trace at
37 the IRQ handler, printing it to the buffer if the *Thread* timer latency is
38 higher than *30 us*. It is also set to stop the session if a *Thread* timer
39 latency higher than *30 us* is hit. Finally, it is set to save the trace
40 buffer if the stop condition is hit::
41
42   [root@alien ~]# rtla timerlat top -s 30 -t 30 -T
43                    Timer Latency
44     0 00:00:59   |          IRQ Timer Latency (us)        |         Thread Timer Latency (us)
45   CPU COUNT      |      cur       min       avg       max |      cur       min       avg       max
46     0 #58634     |        1         0         1        10 |       11         2        10        23
47     1 #58634     |        1         0         1         9 |       12         2         9        23
48     2 #58634     |        0         0         1        11 |       10         2         9        23
49     3 #58634     |        1         0         1        11 |       11         2         9        24
50     4 #58634     |        1         0         1        10 |       11         2         9        26
51     5 #58634     |        1         0         1         8 |       10         2         9        25
52     6 #58634     |       12         0         1        12 |       30         2        10        30 <--- CPU with spike
53     7 #58634     |        1         0         1         9 |       11         2         9        23
54     8 #58633     |        1         0         1         9 |       11         2         9        26
55     9 #58633     |        1         0         1         9 |       10         2         9        26
56    10 #58633     |        1         0         1        13 |       11         2         9        28
57    11 #58633     |        1         0         1        13 |       12         2         9        24
58    12 #58633     |        1         0         1         8 |       10         2         9        23
59    13 #58633     |        1         0         1        10 |       10         2         9        22
60    14 #58633     |        1         0         1        18 |       12         2         9        27
61    15 #58633     |        1         0         1        10 |       11         2         9        28
62    16 #58633     |        0         0         1        11 |        7         2         9        26
63    17 #58633     |        1         0         1        13 |       10         2         9        24
64    18 #58633     |        1         0         1         9 |       13         2         9        22
65    19 #58633     |        1         0         1        10 |       11         2         9        23
66    20 #58633     |        1         0         1        12 |       11         2         9        28
67    21 #58633     |        1         0         1        14 |       11         2         9        24
68    22 #58633     |        1         0         1         8 |       11         2         9        22
69    23 #58633     |        1         0         1        10 |       11         2         9        27
70   timerlat hit stop tracing
71   saving trace to timerlat_trace.txt
72   [root@alien bristot]# tail -60 timerlat_trace.txt
73   [...]
74       timerlat/5-79755   [005] .......   426.271226: #58634 context thread timer_latency     10823 ns
75               sh-109404  [006] dnLh213   426.271247: #58634 context    irq timer_latency     12505 ns
76               sh-109404  [006] dNLh313   426.271258: irq_noise: local_timer:236 start 426.271245463 duration 12553 ns
77               sh-109404  [006] d...313   426.271263: thread_noise:       sh:109404 start 426.271245853 duration 4769 ns
78       timerlat/6-79756   [006] .......   426.271264: #58634 context thread timer_latency     30328 ns
79       timerlat/6-79756   [006] ....1..   426.271265: <stack trace>
80   => timerlat_irq
81   => __hrtimer_run_queues
82   => hrtimer_interrupt
83   => __sysvec_apic_timer_interrupt
84   => sysvec_apic_timer_interrupt
85   => asm_sysvec_apic_timer_interrupt
86   => _raw_spin_unlock_irqrestore                        <---- spinlock that disabled interrupt.
87   => try_to_wake_up
88   => autoremove_wake_function
89   => __wake_up_common
90   => __wake_up_common_lock
91   => ep_poll_callback
92   => __wake_up_common
93   => __wake_up_common_lock
94   => fsnotify_add_event
95   => inotify_handle_inode_event
96   => fsnotify
97   => __fsnotify_parent
98   => __fput
99   => task_work_run
100   => exit_to_user_mode_prepare
101   => syscall_exit_to_user_mode
102   => do_syscall_64
103   => entry_SYSCALL_64_after_hwframe
104   => 0x7265000001378c
105   => 0x10000cea7
106   => 0x25a00000204a
107   => 0x12e302d00000000
108   => 0x19b51010901b6
109   => 0x283ce00726500
110   => 0x61ea308872
111   => 0x00000fe3
112             bash-109109  [007] d..h...   426.271265: #58634 context    irq timer_latency      1211 ns
113       timerlat/6-79756   [006] .......   426.271267: timerlat_main: stop tracing hit on cpu 6
114
115 In the trace, it is possible the notice that the *IRQ* timer latency was
116 already high, accounting *12505 ns*. The IRQ delay was caused by the
117 *bash-109109* process that disabled IRQs in the wake-up path
118 (*_try_to_wake_up()* function). The duration of the IRQ handler that woke
119 up the timerlat thread, informed with the **osnoise:irq_noise** event, was
120 also high and added more *12553 ns* to the Thread latency. Finally, the
121 **osnoise:thread_noise** added by the currently running thread (including
122 the scheduling overhead) added more *4769 ns*. Summing up these values,
123 the *Thread* timer latency accounted for *30328 ns*.
124
125 The primary reason for this high value is the wake-up path that was hit
126 twice during this case: when the *bash-109109* was waking up a thread
127 and then when the *timerlat* thread was awakened. This information can
128 then be used as the starting point of a more fine-grained analysis.
129
130 Note that **rtla timerlat** was dispatched without changing *timerlat* tracer
131 threads' priority. That is generally not needed because these threads hava
132 priority *FIFO:95* by default, which is a common priority used by real-time
133 kernel developers to analyze scheduling delays.
134
135 SEE ALSO
136 --------
137 **rtla-timerlat**\(1), **rtla-timerlat-hist**\(1)
138
139 *timerlat* tracer documentation: <https://www.kernel.org/doc/html/latest/trace/timerlat-tracer.html>
140
141 AUTHOR
142 ------
143 Written by Daniel Bristot de Oliveira <bristot@kernel.org>
144
145 .. include:: common_appendix.rst