rtla-timerlat-top - Man Page
Measures the operating system timer latency
Synopsis
rtla timerlat top [Options] ...
Description
The rtla timerlat tool is an interface for the timerlat tracer. The timerlat tracer dispatches a kernel thread per-cpu. These threads set a periodic timer to wake themselves up and go back to sleep. After the wakeup, they collect and generate useful information for the debugging of operating system timer latency.
The timerlat tracer outputs information in two ways. It periodically prints the timer latency at the timer IRQ handler and the Thread handler. It also enable the trace of the most relevant information via osnoise: tracepoints.
The rtla timerlat top displays a summary of the periodic output from the timerlat tracer. It also provides information for each operating system noise via the osnoise: tracepoints that can be seem with the option -T.
Options
- -a, --auto us
Set the automatic trace mode. This mode sets some commonly used options while debugging the system. It is equivalent to use -T us -s us -t. By default, timerlat tracer uses FIFO:95 for timerlat threads, thus equilavent to -P f:95.
- -p, --period us
Set the timerlat tracer period in microseconds.
- -i, --irq us
Stop trace if the IRQ latency is higher than the argument in us.
- -T, --thread us
Stop trace if the Thread latency is higher than the argument in us.
- -s, --stack us
Save the stack trace at the IRQ if a Thread latency is higher than the argument in us.
- -t, --trace [file]
Save the stopped trace to [file|timerlat_trace.txt].
- --dma-latency us
Set the /dev/cpu_dma_latency to us, aiming to bound exit from idle latencies. cyclictest sets this value to 0 by default, use --dma-latency 0 to have similar results.
- --deepest-idle-state n
Disable idle states higher than n for cpus that are running timerlat threads to reduce exit from idle latencies. If n is -1, all idle states are disabled. On exit from timerlat, the idle state setting is restored to its original state before running timerlat.
Requires rtla to be built with libcpupower.
- -k, --kernel-threads
Use timerlat kernel-space threads, in contrast of -u.
- -u, --user-threads
Set timerlat to run without a workload, and then dispatches user-space workloads to wait on the timerlat_fd. Once the workload is awakes, it goes to sleep again adding so the measurement for the kernel-to-user and user-to-kernel to the tracer output. --user-threads will be used unless the user specify -k.
- -U, --user-load
Set timerlat to run without workload, waiting for the user to dispatch a per-cpu task that waits for a new period on the tracing/osnoise/per_cpu/cpu$ID/timerlat_fd. See linux/tools/rtla/sample/timerlat_load.py for an example of user-load code.
- -q, --quiet
Print only a summary at the end of the session.
- -c, --cpus cpu-list
Set the osnoise tracer to run the sample threads in the cpu-list.
- -H, --house-keeping cpu-list
Run rtla control threads only on the given cpu-list.
- -d, --duration time[s|m|h|d]
Set the duration of the session.
- -D, --debug
Print debug info.
- -e, --event sys:event
Enable an event in the trace (-t) session. The argument can be a specific event, e.g., -e sched:sched_switch, or all events of a system group, e.g., -e sched. Multiple -e are allowed. It is only active when -t or -a are set.
- --filter <filter>
Filter the previous -e sys:event event with <filter>. For further information about event filtering see <https://www.kernel.org/doc/html/latest/trace/events.html#event-filtering> .
- --trigger <trigger>
Enable a trace event trigger to the previous -e sys:event. If the hist: trigger is activated, the output histogram will be automatically saved to a file named system_event_hist.txt. For example, the command:
rtla <command> <mode> -t -e osnoise:irq_noise --trigger="hist:key=desc,duration/1000:sort=desc,duration/1000:vals=hitcount"
Will automatically save the content of the histogram associated to osnoise:irq_noise event in osnoise_irq_noise_hist.txt.
For further information about event trigger see <https://www.kernel.org/doc/html/latest/trace/events.html#event-triggers> .
- -P, --priority o:prio|r:prio|f:prio|d:runtime:period
Set scheduling parameters to the osnoise tracer threads, the format to set the priority are:
- o:prio - use SCHED_OTHER with prio;
- r:prio - use SCHED_RR with prio;
- f:prio - use SCHED_FIFO with prio;
- d:runtime[us|ms|s]:period[us|ms|s] - use SCHED_DEADLINE with runtime and period in nanoseconds.
- -C, --cgroup[=cgroup]
Set a cgroup to the tracer's threads. If the -C option is passed without arguments, the tracer's thread will inherit rtla's cgroup. Otherwise, the threads will be placed on the cgroup passed to the option.
- --warm-up s
After starting the workload, let it run for s seconds before starting collecting the data, allowing the system to warm-up. Statistical data generated during warm-up is discarded.
- --trace-buffer-size kB
Set the per-cpu trace buffer size in kB for the tracing output.
- -h, --help
Print help menu.
- --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
- --aa-only us
Set stop tracing conditions and run without collecting and displaying statistics. Print the auto-analysis if the system hits the stop tracing condition. This option is useful to reduce rtla timerlat CPU, enabling the debug without the overhead of collecting the statistics.
Example
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:
# 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
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.txtIn 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 have priority FIFO:95 by default, which is a common priority used by real-time kernel developers to analyze scheduling delays.
See Also
rtla-timerlat(1), rtla-timerlat-hist(1)
timerlat tracer documentation: < <https://www.kernel.org/doc/html/latest/trace/timerlat-tracer.html> >
Author
Written by Daniel Bristot de Oliveira < <bristot@kernel.org> >
Reporting Bugs
Report bugs to < <linux-kernel@vger.kernel.org> > and < <linux-trace-devel@vger.kernel.org> >
License
rtla is Free Software licensed under the GNU GPLv2
Copying
Copyright (C) 2021 Red Hat, Inc. Free use of this software is granted under the terms of the GNU Public License (GPL).