1 ==========================================
2 Reducing OS jitter due to per-cpu kthreads
3 ==========================================
5 This document lists per-CPU kthreads in the Linux kernel and presents
6 options to control their OS jitter. Note that non-per-CPU kthreads are
7 not listed here. To reduce OS jitter from non-per-CPU kthreads, bind
8 them to a "housekeeping" CPU dedicated to such work.
13 - Documentation/IRQ-affinity.txt: Binding interrupts to sets of CPUs.
15 - Documentation/cgroup-v1: Using cgroups to bind tasks to sets of CPUs.
17 - man taskset: Using the taskset command to bind tasks to sets
20 - man sched_setaffinity: Using the sched_setaffinity() system
21 call to bind tasks to sets of CPUs.
23 - /sys/devices/system/cpu/cpuN/online: Control CPU N's hotplug state,
24 writing "0" to offline and "1" to online.
26 - In order to locate kernel-generated OS jitter on CPU N:
28 cd /sys/kernel/debug/tracing
29 echo 1 > max_graph_depth # Increase the "1" for more detail
30 echo function_graph > current_tracer
32 cat per_cpu/cpuN/trace
41 Periodically process Infiniband-related work.
43 To reduce its OS jitter, do any of the following:
45 1. Don't use eHCA Infiniband hardware, instead choosing hardware
46 that does not require per-CPU kthreads. This will prevent these
47 kthreads from being created in the first place. (This will
48 work for most people, as this hardware, though important, is
49 relatively old and is produced in relatively low unit volumes.)
50 2. Do all eHCA-Infiniband-related work on other CPUs, including
52 3. Rework the eHCA driver so that its per-CPU kthreads are
53 provisioned only on selected CPUs.
60 Handle threaded interrupts.
62 To reduce its OS jitter, do the following:
64 1. Use irq affinity to force the irq threads to execute on
71 Handle Bluetooth work.
73 To reduce its OS jitter, do one of the following:
75 1. Don't use Bluetooth, in which case these kthreads won't be
76 created in the first place.
77 2. Use irq affinity to force Bluetooth-related interrupts to
78 occur on some other CPU and furthermore initiate all
79 Bluetooth activity on some other CPU.
85 Execute softirq handlers when threaded or when under heavy load.
87 To reduce its OS jitter, each softirq vector must be handled
88 separately as follows:
93 Do all of the following:
95 1. To the extent possible, keep the CPU out of the kernel when it
96 is non-idle, for example, by avoiding system calls and by forcing
97 both kernel threads and interrupts to execute elsewhere.
98 2. Build with CONFIG_HOTPLUG_CPU=y. After boot completes, force
99 the CPU offline, then bring it back online. This forces
100 recurring timers to migrate elsewhere. If you are concerned
101 with multiple CPUs, force them all offline before bringing the
102 first one back online. Once you have onlined the CPUs in question,
103 do not offline any other CPUs, because doing so could force the
104 timer back onto one of the CPUs in question.
106 NET_TX_SOFTIRQ and NET_RX_SOFTIRQ
107 ---------------------------------
109 Do all of the following:
111 1. Force networking interrupts onto other CPUs.
112 2. Initiate any network I/O on other CPUs.
113 3. Once your application has started, prevent CPU-hotplug operations
114 from being initiated from tasks that might run on the CPU to
115 be de-jittered. (It is OK to force this CPU offline and then
116 bring it back online before you start your application.)
121 Do all of the following:
123 1. Force block-device interrupts onto some other CPU.
124 2. Initiate any block I/O on other CPUs.
125 3. Once your application has started, prevent CPU-hotplug operations
126 from being initiated from tasks that might run on the CPU to
127 be de-jittered. (It is OK to force this CPU offline and then
128 bring it back online before you start your application.)
133 Do all of the following:
135 1. Force block-device interrupts onto some other CPU.
136 2. Initiate any block I/O and block-I/O polling on other CPUs.
137 3. Once your application has started, prevent CPU-hotplug operations
138 from being initiated from tasks that might run on the CPU to
139 be de-jittered. (It is OK to force this CPU offline and then
140 bring it back online before you start your application.)
145 Do one or more of the following:
147 1. Avoid use of drivers that use tasklets. (Such drivers will contain
148 calls to things like tasklet_schedule().)
149 2. Convert all drivers that you must use from tasklets to workqueues.
150 3. Force interrupts for drivers using tasklets onto other CPUs,
151 and also do I/O involving these drivers on other CPUs.
156 Do all of the following:
158 1. Avoid sending scheduler IPIs to the CPU to be de-jittered,
159 for example, ensure that at most one runnable kthread is present
160 on that CPU. If a thread that expects to run on the de-jittered
161 CPU awakens, the scheduler will send an IPI that can result in
162 a subsequent SCHED_SOFTIRQ.
163 2. CONFIG_NO_HZ_FULL=y and ensure that the CPU to be de-jittered
164 is marked as an adaptive-ticks CPU using the "nohz_full="
165 boot parameter. This reduces the number of scheduler-clock
166 interrupts that the de-jittered CPU receives, minimizing its
167 chances of being selected to do the load balancing work that
168 runs in SCHED_SOFTIRQ context.
169 3. To the extent possible, keep the CPU out of the kernel when it
170 is non-idle, for example, by avoiding system calls and by
171 forcing both kernel threads and interrupts to execute elsewhere.
172 This further reduces the number of scheduler-clock interrupts
173 received by the de-jittered CPU.
178 Do all of the following:
180 1. To the extent possible, keep the CPU out of the kernel when it
181 is non-idle. For example, avoid system calls and force both
182 kernel threads and interrupts to execute elsewhere.
183 2. Build with CONFIG_HOTPLUG_CPU=y. Once boot completes, force the
184 CPU offline, then bring it back online. This forces recurring
185 timers to migrate elsewhere. If you are concerned with multiple
186 CPUs, force them all offline before bringing the first one
187 back online. Once you have onlined the CPUs in question, do not
188 offline any other CPUs, because doing so could force the timer
189 back onto one of the CPUs in question.
194 Do at least one of the following:
196 1. Offload callbacks and keep the CPU in either dyntick-idle or
197 adaptive-ticks state by doing all of the following:
199 a. CONFIG_NO_HZ_FULL=y and ensure that the CPU to be
200 de-jittered is marked as an adaptive-ticks CPU using the
201 "nohz_full=" boot parameter. Bind the rcuo kthreads to
202 housekeeping CPUs, which can tolerate OS jitter.
203 b. To the extent possible, keep the CPU out of the kernel
204 when it is non-idle, for example, by avoiding system
205 calls and by forcing both kernel threads and interrupts
206 to execute elsewhere.
208 2. Enable RCU to do its processing remotely via dyntick-idle by
209 doing all of the following:
211 a. Build with CONFIG_NO_HZ=y and CONFIG_RCU_FAST_NO_HZ=y.
212 b. Ensure that the CPU goes idle frequently, allowing other
213 CPUs to detect that it has passed through an RCU quiescent
214 state. If the kernel is built with CONFIG_NO_HZ_FULL=y,
215 userspace execution also allows other CPUs to detect that
216 the CPU in question has passed through a quiescent state.
217 c. To the extent possible, keep the CPU out of the kernel
218 when it is non-idle, for example, by avoiding system
219 calls and by forcing both kernel threads and interrupts
220 to execute elsewhere.
223 kworker/%u:%d%s (cpu, id, priority)
226 Execute workqueue requests
228 To reduce its OS jitter, do any of the following:
230 1. Run your workload at a real-time priority, which will allow
231 preempting the kworker daemons.
232 2. A given workqueue can be made visible in the sysfs filesystem
233 by passing the WQ_SYSFS to that workqueue's alloc_workqueue().
234 Such a workqueue can be confined to a given subset of the
235 CPUs using the ``/sys/devices/virtual/workqueue/*/cpumask`` sysfs
236 files. The set of WQ_SYSFS workqueues can be displayed using
237 "ls sys/devices/virtual/workqueue". That said, the workqueues
238 maintainer would like to caution people against indiscriminately
239 sprinkling WQ_SYSFS across all the workqueues. The reason for
240 caution is that it is easy to add WQ_SYSFS, but because sysfs is
241 part of the formal user/kernel API, it can be nearly impossible
242 to remove it, even if its addition was a mistake.
243 3. Do any of the following needed to avoid jitter that your
244 application cannot tolerate:
246 a. Build your kernel with CONFIG_SLUB=y rather than
247 CONFIG_SLAB=y, thus avoiding the slab allocator's periodic
248 use of each CPU's workqueues to run its cache_reap()
250 b. Avoid using oprofile, thus avoiding OS jitter from
252 c. Limit your CPU frequency so that a CPU-frequency
253 governor is not required, possibly enlisting the aid of
254 special heatsinks or other cooling technologies. If done
255 correctly, and if you CPU architecture permits, you should
256 be able to build your kernel with CONFIG_CPU_FREQ=n to
257 avoid the CPU-frequency governor periodically running
258 on each CPU, including cs_dbs_timer() and od_dbs_timer().
260 WARNING: Please check your CPU specifications to
261 make sure that this is safe on your particular system.
262 d. As of v3.18, Christoph Lameter's on-demand vmstat workers
263 commit prevents OS jitter due to vmstat_update() on
264 CONFIG_SMP=y systems. Before v3.18, is not possible
265 to entirely get rid of the OS jitter, but you can
266 decrease its frequency by writing a large value to
267 /proc/sys/vm/stat_interval. The default value is HZ,
268 for an interval of one second. Of course, larger values
269 will make your virtual-memory statistics update more
270 slowly. Of course, you can also run your workload at
271 a real-time priority, thus preempting vmstat_update(),
272 but if your workload is CPU-bound, this is a bad idea.
273 However, there is an RFC patch from Christoph Lameter
274 (based on an earlier one from Gilad Ben-Yossef) that
275 reduces or even eliminates vmstat overhead for some
276 workloads at https://lkml.org/lkml/2013/9/4/379.
277 e. Boot with "elevator=noop" to avoid workqueue use by
279 f. If running on high-end powerpc servers, build with
280 CONFIG_PPC_RTAS_DAEMON=n. This prevents the RTAS
281 daemon from running on each CPU every second or so.
282 (This will require editing Kconfig files and will defeat
283 this platform's RAS functionality.) This avoids jitter
284 due to the rtas_event_scan() function.
285 WARNING: Please check your CPU specifications to
286 make sure that this is safe on your particular system.
287 g. If running on Cell Processor, build your kernel with
288 CBE_CPUFREQ_SPU_GOVERNOR=n to avoid OS jitter from
290 WARNING: Please check your CPU specifications to
291 make sure that this is safe on your particular system.
292 h. If running on PowerMAC, build your kernel with
293 CONFIG_PMAC_RACKMETER=n to disable the CPU-meter,
294 avoiding OS jitter from rackmeter_do_timer().
300 Execute RCU callbacks in CONFIG_RCU_BOOST=y kernels.
302 To reduce its OS jitter, do at least one of the following:
304 1. Build the kernel with CONFIG_PREEMPT=n. This prevents these
305 kthreads from being created in the first place, and also obviates
306 the need for RCU priority boosting. This approach is feasible
307 for workloads that do not require high degrees of responsiveness.
308 2. Build the kernel with CONFIG_RCU_BOOST=n. This prevents these
309 kthreads from being created in the first place. This approach
310 is feasible only if your workload never requires RCU priority
311 boosting, for example, if you ensure frequent idle time on all
312 CPUs that might execute within the kernel.
313 3. Build with CONFIG_RCU_NOCB_CPU=y and boot with the rcu_nocbs=
314 boot parameter offloading RCU callbacks from all CPUs susceptible
315 to OS jitter. This approach prevents the rcuc/%u kthreads from
316 having any work to do, so that they are never awakened.
317 4. Ensure that the CPU never enters the kernel, and, in particular,
318 avoid initiating any CPU hotplug operations on this CPU. This is
319 another way of preventing any callbacks from being queued on the
320 CPU, again preventing the rcuc/%u kthreads from having any work
324 rcuob/%d, rcuop/%d, and rcuos/%d
327 Offload RCU callbacks from the corresponding CPU.
329 To reduce its OS jitter, do at least one of the following:
331 1. Use affinity, cgroups, or other mechanism to force these kthreads
332 to execute on some other CPU.
333 2. Build with CONFIG_RCU_NOCB_CPU=n, which will prevent these
334 kthreads from being created in the first place. However, please
335 note that this will not eliminate OS jitter, but will instead
336 shift it to RCU_SOFTIRQ.
342 Detect software lockups on each CPU.
344 To reduce its OS jitter, do at least one of the following:
346 1. Build with CONFIG_LOCKUP_DETECTOR=n, which will prevent these
347 kthreads from being created in the first place.
348 2. Boot with "nosoftlockup=0", which will also prevent these kthreads
349 from being created. Other related watchdog and softlockup boot
350 parameters may be found in Documentation/admin-guide/kernel-parameters.rst
351 and Documentation/watchdog/watchdog-parameters.txt.
352 3. Echo a zero to /proc/sys/kernel/watchdog to disable the
354 4. Echo a large number of /proc/sys/kernel/watchdog_thresh in
355 order to reduce the frequency of OS jitter due to the watchdog
356 timer down to a level that is acceptable for your workload.