Blame view
Documentation/kernel-per-CPU-kthreads.txt
13.3 KB
7d98c21bd kernel-per-CPU-kt... |
1 2 3 |
========================================== Reducing OS jitter due to per-cpu kthreads ========================================== |
49717cb40 kthread: Document... |
4 5 6 7 8 |
This document lists per-CPU kthreads in the Linux kernel and presents options to control their OS jitter. Note that non-per-CPU kthreads are not listed here. To reduce OS jitter from non-per-CPU kthreads, bind them to a "housekeeping" CPU dedicated to such work. |
7d98c21bd kernel-per-CPU-kt... |
9 10 |
References ========== |
49717cb40 kthread: Document... |
11 |
|
7d98c21bd kernel-per-CPU-kt... |
12 |
- Documentation/IRQ-affinity.txt: Binding interrupts to sets of CPUs. |
49717cb40 kthread: Document... |
13 |
|
7d98c21bd kernel-per-CPU-kt... |
14 |
- Documentation/cgroup-v1: Using cgroups to bind tasks to sets of CPUs. |
49717cb40 kthread: Document... |
15 |
|
7d98c21bd kernel-per-CPU-kt... |
16 |
- man taskset: Using the taskset command to bind tasks to sets |
49717cb40 kthread: Document... |
17 |
of CPUs. |
7d98c21bd kernel-per-CPU-kt... |
18 |
- man sched_setaffinity: Using the sched_setaffinity() system |
49717cb40 kthread: Document... |
19 |
call to bind tasks to sets of CPUs. |
7d98c21bd kernel-per-CPU-kt... |
20 |
- /sys/devices/system/cpu/cpuN/online: Control CPU N's hotplug state, |
49717cb40 kthread: Document... |
21 |
writing "0" to offline and "1" to online. |
7d98c21bd kernel-per-CPU-kt... |
22 |
- In order to locate kernel-generated OS jitter on CPU N: |
49717cb40 kthread: Document... |
23 24 25 26 27 28 |
cd /sys/kernel/debug/tracing echo 1 > max_graph_depth # Increase the "1" for more detail echo function_graph > current_tracer # run workload cat per_cpu/cpuN/trace |
7d98c21bd kernel-per-CPU-kt... |
29 30 31 32 33 |
kthreads ======== Name: ehca_comp/%u |
49717cb40 kthread: Document... |
34 |
|
7d98c21bd kernel-per-CPU-kt... |
35 36 |
Purpose: Periodically process Infiniband-related work. |
49717cb40 kthread: Document... |
37 |
|
49717cb40 kthread: Document... |
38 |
To reduce its OS jitter, do any of the following: |
7d98c21bd kernel-per-CPU-kt... |
39 |
|
49717cb40 kthread: Document... |
40 41 42 43 44 45 46 47 48 |
1. Don't use eHCA Infiniband hardware, instead choosing hardware that does not require per-CPU kthreads. This will prevent these kthreads from being created in the first place. (This will work for most people, as this hardware, though important, is relatively old and is produced in relatively low unit volumes.) 2. Do all eHCA-Infiniband-related work on other CPUs, including interrupts. 3. Rework the eHCA driver so that its per-CPU kthreads are provisioned only on selected CPUs. |
7d98c21bd kernel-per-CPU-kt... |
49 50 51 52 53 |
Name: irq/%d-%s Purpose: Handle threaded interrupts. |
49717cb40 kthread: Document... |
54 |
To reduce its OS jitter, do the following: |
7d98c21bd kernel-per-CPU-kt... |
55 |
|
49717cb40 kthread: Document... |
56 57 |
1. Use irq affinity to force the irq threads to execute on some other CPU. |
7d98c21bd kernel-per-CPU-kt... |
58 59 60 61 62 |
Name: kcmtpd_ctr_%d Purpose: Handle Bluetooth work. |
49717cb40 kthread: Document... |
63 |
To reduce its OS jitter, do one of the following: |
7d98c21bd kernel-per-CPU-kt... |
64 |
|
49717cb40 kthread: Document... |
65 66 67 68 69 |
1. Don't use Bluetooth, in which case these kthreads won't be created in the first place. 2. Use irq affinity to force Bluetooth-related interrupts to occur on some other CPU and furthermore initiate all Bluetooth activity on some other CPU. |
7d98c21bd kernel-per-CPU-kt... |
70 71 72 73 74 |
Name: ksoftirqd/%u Purpose: Execute softirq handlers when threaded or when under heavy load. |
49717cb40 kthread: Document... |
75 76 |
To reduce its OS jitter, each softirq vector must be handled separately as follows: |
7d98c21bd kernel-per-CPU-kt... |
77 78 79 80 81 |
TIMER_SOFTIRQ ------------- Do all of the following: |
49717cb40 kthread: Document... |
82 83 84 85 86 87 88 89 90 91 |
1. To the extent possible, keep the CPU out of the kernel when it is non-idle, for example, by avoiding system calls and by forcing both kernel threads and interrupts to execute elsewhere. 2. Build with CONFIG_HOTPLUG_CPU=y. After boot completes, force the CPU offline, then bring it back online. This forces recurring timers to migrate elsewhere. If you are concerned with multiple CPUs, force them all offline before bringing the first one back online. Once you have onlined the CPUs in question, do not offline any other CPUs, because doing so could force the timer back onto one of the CPUs in question. |
7d98c21bd kernel-per-CPU-kt... |
92 93 94 95 96 |
NET_TX_SOFTIRQ and NET_RX_SOFTIRQ --------------------------------- Do all of the following: |
49717cb40 kthread: Document... |
97 98 99 100 101 102 |
1. Force networking interrupts onto other CPUs. 2. Initiate any network I/O on other CPUs. 3. Once your application has started, prevent CPU-hotplug operations from being initiated from tasks that might run on the CPU to be de-jittered. (It is OK to force this CPU offline and then bring it back online before you start your application.) |
7d98c21bd kernel-per-CPU-kt... |
103 104 105 106 107 |
BLOCK_SOFTIRQ ------------- Do all of the following: |
49717cb40 kthread: Document... |
108 109 110 111 112 113 |
1. Force block-device interrupts onto some other CPU. 2. Initiate any block I/O on other CPUs. 3. Once your application has started, prevent CPU-hotplug operations from being initiated from tasks that might run on the CPU to be de-jittered. (It is OK to force this CPU offline and then bring it back online before you start your application.) |
7d98c21bd kernel-per-CPU-kt... |
114 115 116 117 118 |
IRQ_POLL_SOFTIRQ ---------------- Do all of the following: |
49717cb40 kthread: Document... |
119 120 121 122 123 124 |
1. Force block-device interrupts onto some other CPU. 2. Initiate any block I/O and block-I/O polling on other CPUs. 3. Once your application has started, prevent CPU-hotplug operations from being initiated from tasks that might run on the CPU to be de-jittered. (It is OK to force this CPU offline and then bring it back online before you start your application.) |
7d98c21bd kernel-per-CPU-kt... |
125 126 127 128 129 |
TASKLET_SOFTIRQ --------------- Do one or more of the following: |
49717cb40 kthread: Document... |
130 131 132 133 134 |
1. Avoid use of drivers that use tasklets. (Such drivers will contain calls to things like tasklet_schedule().) 2. Convert all drivers that you must use from tasklets to workqueues. 3. Force interrupts for drivers using tasklets onto other CPUs, and also do I/O involving these drivers on other CPUs. |
7d98c21bd kernel-per-CPU-kt... |
135 136 137 138 139 |
SCHED_SOFTIRQ ------------- Do all of the following: |
49717cb40 kthread: Document... |
140 141 142 143 144 |
1. Avoid sending scheduler IPIs to the CPU to be de-jittered, for example, ensure that at most one runnable kthread is present on that CPU. If a thread that expects to run on the de-jittered CPU awakens, the scheduler will send an IPI that can result in a subsequent SCHED_SOFTIRQ. |
44c65ff2e rcu: Eliminate NO... |
145 146 147 148 149 150 |
2. CONFIG_NO_HZ_FULL=y and ensure that the CPU to be de-jittered is marked as an adaptive-ticks CPU using the "nohz_full=" boot parameter. This reduces the number of scheduler-clock interrupts that the de-jittered CPU receives, minimizing its chances of being selected to do the load balancing work that runs in SCHED_SOFTIRQ context. |
49717cb40 kthread: Document... |
151 152 153 154 155 |
3. To the extent possible, keep the CPU out of the kernel when it is non-idle, for example, by avoiding system calls and by forcing both kernel threads and interrupts to execute elsewhere. This further reduces the number of scheduler-clock interrupts received by the de-jittered CPU. |
7d98c21bd kernel-per-CPU-kt... |
156 157 158 159 160 |
HRTIMER_SOFTIRQ --------------- Do all of the following: |
49717cb40 kthread: Document... |
161 162 163 164 165 166 167 168 169 170 |
1. To the extent possible, keep the CPU out of the kernel when it is non-idle. For example, avoid system calls and force both kernel threads and interrupts to execute elsewhere. 2. Build with CONFIG_HOTPLUG_CPU=y. Once boot completes, force the CPU offline, then bring it back online. This forces recurring timers to migrate elsewhere. If you are concerned with multiple CPUs, force them all offline before bringing the first one back online. Once you have onlined the CPUs in question, do not offline any other CPUs, because doing so could force the timer back onto one of the CPUs in question. |
7d98c21bd kernel-per-CPU-kt... |
171 172 173 174 175 |
RCU_SOFTIRQ ----------- Do at least one of the following: |
49717cb40 kthread: Document... |
176 177 |
1. Offload callbacks and keep the CPU in either dyntick-idle or adaptive-ticks state by doing all of the following: |
7d98c21bd kernel-per-CPU-kt... |
178 |
|
44c65ff2e rcu: Eliminate NO... |
179 180 181 182 |
a. CONFIG_NO_HZ_FULL=y and ensure that the CPU to be de-jittered is marked as an adaptive-ticks CPU using the "nohz_full=" boot parameter. Bind the rcuo kthreads to housekeeping CPUs, which can tolerate OS jitter. |
49717cb40 kthread: Document... |
183 184 185 186 |
b. To the extent possible, keep the CPU out of the kernel when it is non-idle, for example, by avoiding system calls and by forcing both kernel threads and interrupts to execute elsewhere. |
7d98c21bd kernel-per-CPU-kt... |
187 |
|
49717cb40 kthread: Document... |
188 189 |
2. Enable RCU to do its processing remotely via dyntick-idle by doing all of the following: |
7d98c21bd kernel-per-CPU-kt... |
190 |
|
49717cb40 kthread: Document... |
191 192 193 194 195 196 197 198 199 200 |
a. Build with CONFIG_NO_HZ=y and CONFIG_RCU_FAST_NO_HZ=y. b. Ensure that the CPU goes idle frequently, allowing other CPUs to detect that it has passed through an RCU quiescent state. If the kernel is built with CONFIG_NO_HZ_FULL=y, userspace execution also allows other CPUs to detect that the CPU in question has passed through a quiescent state. c. To the extent possible, keep the CPU out of the kernel when it is non-idle, for example, by avoiding system calls and by forcing both kernel threads and interrupts to execute elsewhere. |
7d98c21bd kernel-per-CPU-kt... |
201 202 203 204 205 |
Name: kworker/%u:%d%s (cpu, id, priority) Purpose: Execute workqueue requests |
f7bac9b85 kthread: Add kwor... |
206 |
To reduce its OS jitter, do any of the following: |
7d98c21bd kernel-per-CPU-kt... |
207 |
|
f7bac9b85 kthread: Add kwor... |
208 209 |
1. Run your workload at a real-time priority, which will allow preempting the kworker daemons. |
bbf393b0d Documentation/ker... |
210 211 212 |
2. A given workqueue can be made visible in the sysfs filesystem by passing the WQ_SYSFS to that workqueue's alloc_workqueue(). Such a workqueue can be confined to a given subset of the |
7d98c21bd kernel-per-CPU-kt... |
213 |
CPUs using the ``/sys/devices/virtual/workqueue/*/cpumask`` sysfs |
bbf393b0d Documentation/ker... |
214 215 216 217 218 219 220 221 |
files. The set of WQ_SYSFS workqueues can be displayed using "ls sys/devices/virtual/workqueue". That said, the workqueues maintainer would like to caution people against indiscriminately sprinkling WQ_SYSFS across all the workqueues. The reason for caution is that it is easy to add WQ_SYSFS, but because sysfs is part of the formal user/kernel API, it can be nearly impossible to remove it, even if its addition was a mistake. 3. Do any of the following needed to avoid jitter that your |
f7bac9b85 kthread: Add kwor... |
222 |
application cannot tolerate: |
7d98c21bd kernel-per-CPU-kt... |
223 |
|
f7bac9b85 kthread: Add kwor... |
224 225 226 227 228 229 230 231 232 233 234 235 236 |
a. Build your kernel with CONFIG_SLUB=y rather than CONFIG_SLAB=y, thus avoiding the slab allocator's periodic use of each CPU's workqueues to run its cache_reap() function. b. Avoid using oprofile, thus avoiding OS jitter from wq_sync_buffer(). c. Limit your CPU frequency so that a CPU-frequency governor is not required, possibly enlisting the aid of special heatsinks or other cooling technologies. If done correctly, and if you CPU architecture permits, you should be able to build your kernel with CONFIG_CPU_FREQ=n to avoid the CPU-frequency governor periodically running on each CPU, including cs_dbs_timer() and od_dbs_timer(). |
7d98c21bd kernel-per-CPU-kt... |
237 |
|
f7bac9b85 kthread: Add kwor... |
238 239 |
WARNING: Please check your CPU specifications to make sure that this is safe on your particular system. |
89bf5d82e documentation: Up... |
240 241 242 243 244 245 246 247 248 249 |
d. As of v3.18, Christoph Lameter's on-demand vmstat workers commit prevents OS jitter due to vmstat_update() on CONFIG_SMP=y systems. Before v3.18, is not possible to entirely get rid of the OS jitter, but you can decrease its frequency by writing a large value to /proc/sys/vm/stat_interval. The default value is HZ, for an interval of one second. Of course, larger values will make your virtual-memory statistics update more slowly. Of course, you can also run your workload at a real-time priority, thus preempting vmstat_update(), |
64f26e5c8 kthread: Add poin... |
250 251 252 253 254 |
but if your workload is CPU-bound, this is a bad idea. However, there is an RFC patch from Christoph Lameter (based on an earlier one from Gilad Ben-Yossef) that reduces or even eliminates vmstat overhead for some workloads at https://lkml.org/lkml/2013/9/4/379. |
f1360570f documentation: Up... |
255 256 257 |
e. Boot with "elevator=noop" to avoid workqueue use by the block layer. f. If running on high-end powerpc servers, build with |
f7bac9b85 kthread: Add kwor... |
258 259 260 261 262 263 264 |
CONFIG_PPC_RTAS_DAEMON=n. This prevents the RTAS daemon from running on each CPU every second or so. (This will require editing Kconfig files and will defeat this platform's RAS functionality.) This avoids jitter due to the rtas_event_scan() function. WARNING: Please check your CPU specifications to make sure that this is safe on your particular system. |
f1360570f documentation: Up... |
265 |
g. If running on Cell Processor, build your kernel with |
f7bac9b85 kthread: Add kwor... |
266 267 268 269 |
CBE_CPUFREQ_SPU_GOVERNOR=n to avoid OS jitter from spu_gov_work(). WARNING: Please check your CPU specifications to make sure that this is safe on your particular system. |
f1360570f documentation: Up... |
270 |
h. If running on PowerMAC, build your kernel with |
f7bac9b85 kthread: Add kwor... |
271 272 |
CONFIG_PMAC_RACKMETER=n to disable the CPU-meter, avoiding OS jitter from rackmeter_do_timer(). |
7d98c21bd kernel-per-CPU-kt... |
273 274 275 276 277 |
Name: rcuc/%u Purpose: Execute RCU callbacks in CONFIG_RCU_BOOST=y kernels. |
49717cb40 kthread: Document... |
278 |
To reduce its OS jitter, do at least one of the following: |
7d98c21bd kernel-per-CPU-kt... |
279 |
|
49717cb40 kthread: Document... |
280 281 282 283 284 285 286 287 288 |
1. Build the kernel with CONFIG_PREEMPT=n. This prevents these kthreads from being created in the first place, and also obviates the need for RCU priority boosting. This approach is feasible for workloads that do not require high degrees of responsiveness. 2. Build the kernel with CONFIG_RCU_BOOST=n. This prevents these kthreads from being created in the first place. This approach is feasible only if your workload never requires RCU priority boosting, for example, if you ensure frequent idle time on all CPUs that might execute within the kernel. |
44c65ff2e rcu: Eliminate NO... |
289 290 291 292 |
3. Build with CONFIG_RCU_NOCB_CPU=y and boot with the rcu_nocbs= boot parameter offloading RCU callbacks from all CPUs susceptible to OS jitter. This approach prevents the rcuc/%u kthreads from having any work to do, so that they are never awakened. |
49717cb40 kthread: Document... |
293 294 295 296 297 |
4. Ensure that the CPU never enters the kernel, and, in particular, avoid initiating any CPU hotplug operations on this CPU. This is another way of preventing any callbacks from being queued on the CPU, again preventing the rcuc/%u kthreads from having any work to do. |
7d98c21bd kernel-per-CPU-kt... |
298 299 300 301 302 |
Name: rcuob/%d, rcuop/%d, and rcuos/%d Purpose: Offload RCU callbacks from the corresponding CPU. |
49717cb40 kthread: Document... |
303 |
To reduce its OS jitter, do at least one of the following: |
7d98c21bd kernel-per-CPU-kt... |
304 |
|
49717cb40 kthread: Document... |
305 306 |
1. Use affinity, cgroups, or other mechanism to force these kthreads to execute on some other CPU. |
b96516220 rcu: Fix Document... |
307 |
2. Build with CONFIG_RCU_NOCB_CPU=n, which will prevent these |
49717cb40 kthread: Document... |
308 309 310 |
kthreads from being created in the first place. However, please note that this will not eliminate OS jitter, but will instead shift it to RCU_SOFTIRQ. |
7d98c21bd kernel-per-CPU-kt... |
311 312 313 314 315 |
Name: watchdog/%u Purpose: Detect software lockups on each CPU. |
49717cb40 kthread: Document... |
316 |
To reduce its OS jitter, do at least one of the following: |
7d98c21bd kernel-per-CPU-kt... |
317 |
|
49717cb40 kthread: Document... |
318 319 |
1. Build with CONFIG_LOCKUP_DETECTOR=n, which will prevent these kthreads from being created in the first place. |
f1360570f documentation: Up... |
320 321 |
2. Boot with "nosoftlockup=0", which will also prevent these kthreads from being created. Other related watchdog and softlockup boot |
8c27ceff3 docs: fix locatio... |
322 |
parameters may be found in Documentation/admin-guide/kernel-parameters.rst |
f1360570f documentation: Up... |
323 324 |
and Documentation/watchdog/watchdog-parameters.txt. 3. Echo a zero to /proc/sys/kernel/watchdog to disable the |
49717cb40 kthread: Document... |
325 |
watchdog timer. |
f1360570f documentation: Up... |
326 |
4. Echo a large number of /proc/sys/kernel/watchdog_thresh in |
49717cb40 kthread: Document... |
327 328 |
order to reduce the frequency of OS jitter due to the watchdog timer down to a level that is acceptable for your workload. |