diff options
Diffstat (limited to 'Documentation/tools')
| -rw-r--r-- | Documentation/tools/rtla/common_timerlat_aa.rst | 7 | ||||
| -rw-r--r-- | Documentation/tools/rtla/index.rst | 1 | ||||
| -rw-r--r-- | Documentation/tools/rtla/rtla-hwnoise.rst | 107 | ||||
| -rw-r--r-- | Documentation/tools/rtla/rtla-timerlat-top.rst | 164 |
4 files changed, 188 insertions, 91 deletions
diff --git a/Documentation/tools/rtla/common_timerlat_aa.rst b/Documentation/tools/rtla/common_timerlat_aa.rst new file mode 100644 index 000000000000..077029e6b289 --- /dev/null +++ b/Documentation/tools/rtla/common_timerlat_aa.rst @@ -0,0 +1,7 @@ +**--dump-tasks** + + prints the task running on all CPUs if stop conditions are met (depends on !--no-aa) + +**--no-aa** + + disable auto-analysis, reducing rtla timerlat cpu usage diff --git a/Documentation/tools/rtla/index.rst b/Documentation/tools/rtla/index.rst index 840f0bf3e803..05d2652e4072 100644 --- a/Documentation/tools/rtla/index.rst +++ b/Documentation/tools/rtla/index.rst @@ -17,6 +17,7 @@ behavior on specific hardware. rtla-timerlat rtla-timerlat-hist rtla-timerlat-top + rtla-hwnoise .. only:: subproject and html diff --git a/Documentation/tools/rtla/rtla-hwnoise.rst b/Documentation/tools/rtla/rtla-hwnoise.rst new file mode 100644 index 000000000000..fb1c52bbc00b --- /dev/null +++ b/Documentation/tools/rtla/rtla-hwnoise.rst @@ -0,0 +1,107 @@ +.. SPDX-License-Identifier: GPL-2.0 + +============ +rtla-hwnoise +============ +------------------------------------------ +Detect and quantify hardware-related noise +------------------------------------------ + +:Manual section: 1 + +SYNOPSIS +======== + +**rtla hwnoise** [*OPTIONS*] + +DESCRIPTION +=========== + +**rtla hwnoise** collects the periodic summary from the *osnoise* tracer +running with *interrupts disabled*. By disabling interrupts, and the scheduling +of threads as a consequence, only non-maskable interrupts and hardware-related +noise is allowed. + +The tool also allows the configurations of the *osnoise* tracer and the +collection of the tracer output. + +OPTIONS +======= +.. include:: common_osnoise_options.rst + +.. include:: common_top_options.rst + +.. include:: common_options.rst + +EXAMPLE +======= +In the example below, the **rtla hwnoise** tool is set to run on CPUs *1-7* +on a system with 8 cores/16 threads with hyper-threading enabled. + +The tool is set to detect any noise higher than *one microsecond*, +to run for *ten minutes*, displaying a summary of the report at the +end of the session:: + + # rtla hwnoise -c 1-7 -T 1 -d 10m -q + Hardware-related Noise + duration: 0 00:10:00 | time is in us + CPU Period Runtime Noise % CPU Aval Max Noise Max Single HW NMI + 1 #599 599000000 138 99.99997 3 3 4 74 + 2 #599 599000000 85 99.99998 3 3 4 75 + 3 #599 599000000 86 99.99998 4 3 6 75 + 4 #599 599000000 81 99.99998 4 4 2 75 + 5 #599 599000000 85 99.99998 2 2 2 75 + 6 #599 599000000 76 99.99998 2 2 0 75 + 7 #599 599000000 77 99.99998 3 3 0 75 + + +The first column shows the *CPU*, and the second column shows how many +*Periods* the tool ran during the session. The *Runtime* is the time +the tool effectively runs on the CPU. The *Noise* column is the sum of +all noise that the tool observed, and the *% CPU Aval* is the relation +between the *Runtime* and *Noise*. + +The *Max Noise* column is the maximum hardware noise the tool detected in a +single period, and the *Max Single* is the maximum single noise seen. + +The *HW* and *NMI* columns show the total number of *hardware* and *NMI* noise +occurrence observed by the tool. + +For example, *CPU 3* ran *599* periods of *1 second Runtime*. The CPU received +*86 us* of noise during the entire execution, leaving *99.99997 %* of CPU time +for the application. In the worst single period, the CPU caused *4 us* of +noise to the application, but it was certainly caused by more than one single +noise, as the *Max Single* noise was of *3 us*. The CPU has *HW noise,* at a +rate of *six occurrences*/*ten minutes*. The CPU also has *NMIs*, at a higher +frequency: around *seven per second*. + +The tool should report *0* hardware-related noise in the ideal situation. +For example, by disabling hyper-threading to remove the hardware noise, +and disabling the TSC watchdog to remove the NMI (it is possible to identify +this using tracing options of **rtla hwnoise**), it was possible to reach +the ideal situation in the same hardware:: + + # rtla hwnoise -c 1-7 -T 1 -d 10m -q + Hardware-related Noise + duration: 0 00:10:00 | time is in us + CPU Period Runtime Noise % CPU Aval Max Noise Max Single HW NMI + 1 #599 599000000 0 100.00000 0 0 0 0 + 2 #599 599000000 0 100.00000 0 0 0 0 + 3 #599 599000000 0 100.00000 0 0 0 0 + 4 #599 599000000 0 100.00000 0 0 0 0 + 5 #599 599000000 0 100.00000 0 0 0 0 + 6 #599 599000000 0 100.00000 0 0 0 0 + 7 #599 599000000 0 100.00000 0 0 0 0 + +SEE ALSO +======== + +**rtla-osnoise**\(1) + +Osnoise tracer documentation: <https://www.kernel.org/doc/html/latest/trace/osnoise-tracer.html> + +AUTHOR +====== +Written by Daniel Bristot de Oliveira <bristot@kernel.org> + +.. include:: common_appendix.rst diff --git a/Documentation/tools/rtla/rtla-timerlat-top.rst b/Documentation/tools/rtla/rtla-timerlat-top.rst index 7c4e4b109493..73799c1150ad 100644 --- a/Documentation/tools/rtla/rtla-timerlat-top.rst +++ b/Documentation/tools/rtla/rtla-timerlat-top.rst @@ -30,102 +30,84 @@ OPTIONS .. include:: common_options.rst +.. include:: common_timerlat_aa.rst + EXAMPLE ======= -In the example below, the *timerlat* tracer is set to capture the stack trace at -the IRQ handler, printing it to the buffer if the *Thread* timer latency is -higher than *30 us*. It is also set to stop the session if a *Thread* timer -latency higher than *30 us* is hit. Finally, it is set to save the trace -buffer if the stop condition is hit:: +In the example below, the timerlat tracer is dispatched in cpus *1-23* in the +automatic trace mode, instructing the tracer to stop if a *40 us* latency or +higher is found:: - [root@alien ~]# rtla timerlat top -s 30 -T 30 -t - Timer Latency - 0 00:00:59 | IRQ Timer Latency (us) | Thread Timer Latency (us) + # timerlat -a 40 -c 1-23 -q + Timer Latency + 0 00:00:12 | IRQ Timer Latency (us) | Thread Timer Latency (us) CPU COUNT | cur min avg max | cur min avg max - 0 #58634 | 1 0 1 10 | 11 2 10 23 - 1 #58634 | 1 0 1 9 | 12 2 9 23 - 2 #58634 | 0 0 1 11 | 10 2 9 23 - 3 #58634 | 1 0 1 11 | 11 2 9 24 - 4 #58634 | 1 0 1 10 | 11 2 9 26 - 5 #58634 | 1 0 1 8 | 10 2 9 25 - 6 #58634 | 12 0 1 12 | 30 2 10 30 <--- CPU with spike - 7 #58634 | 1 0 1 9 | 11 2 9 23 - 8 #58633 | 1 0 1 9 | 11 2 9 26 - 9 #58633 | 1 0 1 9 | 10 2 9 26 - 10 #58633 | 1 0 1 13 | 11 2 9 28 - 11 #58633 | 1 0 1 13 | 12 2 9 24 - 12 #58633 | 1 0 1 8 | 10 2 9 23 - 13 #58633 | 1 0 1 10 | 10 2 9 22 - 14 #58633 | 1 0 1 18 | 12 2 9 27 - 15 #58633 | 1 0 1 10 | 11 2 9 28 - 16 #58633 | 0 0 1 11 | 7 2 9 26 - 17 #58633 | 1 0 1 13 | 10 2 9 24 - 18 #58633 | 1 0 1 9 | 13 2 9 22 - 19 #58633 | 1 0 1 10 | 11 2 9 23 - 20 #58633 | 1 0 1 12 | 11 2 9 28 - 21 #58633 | 1 0 1 14 | 11 2 9 24 - 22 #58633 | 1 0 1 8 | 11 2 9 22 - 23 #58633 | 1 0 1 10 | 11 2 9 27 - timerlat hit stop tracing - saving trace to timerlat_trace.txt - [root@alien bristot]# tail -60 timerlat_trace.txt - [...] - timerlat/5-79755 [005] ....... 426.271226: #58634 context thread timer_latency 10823 ns - sh-109404 [006] dnLh213 426.271247: #58634 context irq timer_latency 12505 ns - sh-109404 [006] dNLh313 426.271258: irq_noise: local_timer:236 start 426.271245463 duration 12553 ns - sh-109404 [006] d...313 426.271263: thread_noise: sh:109404 start 426.271245853 duration 4769 ns - timerlat/6-79756 [006] ....... 426.271264: #58634 context thread timer_latency 30328 ns - timerlat/6-79756 [006] ....1.. 426.271265: <stack trace> - => timerlat_irq - => __hrtimer_run_queues - => hrtimer_interrupt - => __sysvec_apic_timer_interrupt - => sysvec_apic_timer_interrupt - => asm_sysvec_apic_timer_interrupt - => _raw_spin_unlock_irqrestore <---- spinlock that disabled interrupt. - => try_to_wake_up - => autoremove_wake_function - => __wake_up_common - => __wake_up_common_lock - => ep_poll_callback - => __wake_up_common - => __wake_up_common_lock - => fsnotify_add_event - => inotify_handle_inode_event - => fsnotify - => __fsnotify_parent - => __fput - => task_work_run - => exit_to_user_mode_prepare - => syscall_exit_to_user_mode - => do_syscall_64 - => entry_SYSCALL_64_after_hwframe - => 0x7265000001378c - => 0x10000cea7 - => 0x25a00000204a - => 0x12e302d00000000 - => 0x19b51010901b6 - => 0x283ce00726500 - => 0x61ea308872 - => 0x00000fe3 - bash-109109 [007] d..h... 426.271265: #58634 context irq timer_latency 1211 ns - timerlat/6-79756 [006] ....... 426.271267: timerlat_main: stop tracing hit on cpu 6 - -In the trace, it is possible the notice that the *IRQ* timer latency was -already high, accounting *12505 ns*. The IRQ delay was caused by the -*bash-109109* process that disabled IRQs in the wake-up path -(*_try_to_wake_up()* function). The duration of the IRQ handler that woke -up the timerlat thread, informed with the **osnoise:irq_noise** event, was -also high and added more *12553 ns* to the Thread latency. Finally, the -**osnoise:thread_noise** added by the currently running thread (including -the scheduling overhead) added more *4769 ns*. Summing up these values, -the *Thread* timer latency accounted for *30328 ns*. - -The primary reason for this high value is the wake-up path that was hit -twice during this case: when the *bash-109109* was waking up a thread -and then when the *timerlat* thread was awakened. This information can -then be used as the starting point of a more fine-grained analysis. + 1 #12322 | 0 0 1 15 | 10 3 9 31 + 2 #12322 | 3 0 1 12 | 10 3 9 23 + 3 #12322 | 1 0 1 21 | 8 2 8 34 + 4 #12322 | 1 0 1 17 | 10 2 11 33 + 5 #12322 | 0 0 1 12 | 8 3 8 25 + 6 #12322 | 1 0 1 14 | 16 3 11 35 + 7 #12322 | 0 0 1 14 | 9 2 8 29 + 8 #12322 | 1 0 1 22 | 9 3 9 34 + 9 #12322 | 0 0 1 14 | 8 2 8 24 + 10 #12322 | 1 0 0 12 | 9 3 8 24 + 11 #12322 | 0 0 0 15 | 6 2 7 29 + 12 #12321 | 1 0 0 13 | 5 3 8 23 + 13 #12319 | 0 0 1 14 | 9 3 9 26 + 14 #12321 | 1 0 0 13 | 6 2 8 24 + 15 #12321 | 1 0 1 15 | 12 3 11 27 + 16 #12318 | 0 0 1 13 | 7 3 10 24 + 17 #12319 | 0 0 1 13 | 11 3 9 25 + 18 #12318 | 0 0 0 12 | 8 2 8 20 + 19 #12319 | 0 0 1 18 | 10 2 9 28 + 20 #12317 | 0 0 0 20 | 9 3 8 34 + 21 #12318 | 0 0 0 13 | 8 3 8 28 + 22 #12319 | 0 0 1 11 | 8 3 10 22 + 23 #12320 | 28 0 1 28 | 41 3 11 41 + rtla timerlat hit stop tracing + ## CPU 23 hit stop tracing, analyzing it ## + IRQ handler delay: 27.49 us (65.52 %) + IRQ latency: 28.13 us + Timerlat IRQ duration: 9.59 us (22.85 %) + Blocking thread: 3.79 us (9.03 %) + objtool:49256 3.79 us + Blocking thread stacktrace + -> timerlat_irq + -> __hrtimer_run_queues + -> hrtimer_interrupt + -> __sysvec_apic_timer_interrupt + -> sysvec_apic_timer_interrupt + -> asm_sysvec_apic_timer_interrupt + -> _raw_spin_unlock_irqrestore + -> cgroup_rstat_flush_locked + -> cgroup_rstat_flush_irqsafe + -> mem_cgroup_flush_stats + -> mem_cgroup_wb_stats + -> balance_dirty_pages + -> balance_dirty_pages_ratelimited_flags + -> btrfs_buffered_write + -> btrfs_do_write_iter + -> vfs_write + -> __x64_sys_pwrite64 + -> do_syscall_64 + -> entry_SYSCALL_64_after_hwframe + ------------------------------------------------------------------------ + Thread latency: 41.96 us (100%) + + The system has exit from idle latency! + Max timerlat IRQ latency from idle: 17.48 us in cpu 4 + Saving trace to timerlat_trace.txt + +In this case, the major factor was the delay suffered by the *IRQ handler* +that handles **timerlat** wakeup: *65.52%*. This can be caused by the +current thread masking interrupts, which can be seen in the blocking +thread stacktrace: the current thread (*objtool:49256*) disabled interrupts +via *raw spin lock* operations inside mem cgroup, while doing write +syscall in a btrfs file system. + +The raw trace is saved in the **timerlat_trace.txt** file for further analysis. Note that **rtla timerlat** was dispatched without changing *timerlat* tracer threads' priority. That is generally not needed because these threads hava |