stallwarn.rst 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336
  1. .. SPDX-License-Identifier: GPL-2.0
  2. ==============================
  3. Using RCU's CPU Stall Detector
  4. ==============================
  5. This document first discusses what sorts of issues RCU's CPU stall
  6. detector can locate, and then discusses kernel parameters and Kconfig
  7. options that can be used to fine-tune the detector's operation. Finally,
  8. this document explains the stall detector's "splat" format.
  9. What Causes RCU CPU Stall Warnings?
  10. ===================================
  11. So your kernel printed an RCU CPU stall warning. The next question is
  12. "What caused it?" The following problems can result in RCU CPU stall
  13. warnings:
  14. - A CPU looping in an RCU read-side critical section.
  15. - A CPU looping with interrupts disabled.
  16. - A CPU looping with preemption disabled.
  17. - A CPU looping with bottom halves disabled.
  18. - For !CONFIG_PREEMPT kernels, a CPU looping anywhere in the kernel
  19. without invoking schedule(). If the looping in the kernel is
  20. really expected and desirable behavior, you might need to add
  21. some calls to cond_resched().
  22. - Booting Linux using a console connection that is too slow to
  23. keep up with the boot-time console-message rate. For example,
  24. a 115Kbaud serial console can be -way- too slow to keep up
  25. with boot-time message rates, and will frequently result in
  26. RCU CPU stall warning messages. Especially if you have added
  27. debug printk()s.
  28. - Anything that prevents RCU's grace-period kthreads from running.
  29. This can result in the "All QSes seen" console-log message.
  30. This message will include information on when the kthread last
  31. ran and how often it should be expected to run. It can also
  32. result in the ``rcu_.*kthread starved for`` console-log message,
  33. which will include additional debugging information.
  34. - A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might
  35. happen to preempt a low-priority task in the middle of an RCU
  36. read-side critical section. This is especially damaging if
  37. that low-priority task is not permitted to run on any other CPU,
  38. in which case the next RCU grace period can never complete, which
  39. will eventually cause the system to run out of memory and hang.
  40. While the system is in the process of running itself out of
  41. memory, you might see stall-warning messages.
  42. - A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that
  43. is running at a higher priority than the RCU softirq threads.
  44. This will prevent RCU callbacks from ever being invoked,
  45. and in a CONFIG_PREEMPT_RCU kernel will further prevent
  46. RCU grace periods from ever completing. Either way, the
  47. system will eventually run out of memory and hang. In the
  48. CONFIG_PREEMPT_RCU case, you might see stall-warning
  49. messages.
  50. You can use the rcutree.kthread_prio kernel boot parameter to
  51. increase the scheduling priority of RCU's kthreads, which can
  52. help avoid this problem. However, please note that doing this
  53. can increase your system's context-switch rate and thus degrade
  54. performance.
  55. - A periodic interrupt whose handler takes longer than the time
  56. interval between successive pairs of interrupts. This can
  57. prevent RCU's kthreads and softirq handlers from running.
  58. Note that certain high-overhead debugging options, for example
  59. the function_graph tracer, can result in interrupt handler taking
  60. considerably longer than normal, which can in turn result in
  61. RCU CPU stall warnings.
  62. - Testing a workload on a fast system, tuning the stall-warning
  63. timeout down to just barely avoid RCU CPU stall warnings, and then
  64. running the same workload with the same stall-warning timeout on a
  65. slow system. Note that thermal throttling and on-demand governors
  66. can cause a single system to be sometimes fast and sometimes slow!
  67. - A hardware or software issue shuts off the scheduler-clock
  68. interrupt on a CPU that is not in dyntick-idle mode. This
  69. problem really has happened, and seems to be most likely to
  70. result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels.
  71. - A hardware or software issue that prevents time-based wakeups
  72. from occurring. These issues can range from misconfigured or
  73. buggy timer hardware through bugs in the interrupt or exception
  74. path (whether hardware, firmware, or software) through bugs
  75. in Linux's timer subsystem through bugs in the scheduler, and,
  76. yes, even including bugs in RCU itself.
  77. - A bug in the RCU implementation.
  78. - A hardware failure. This is quite unlikely, but has occurred
  79. at least once in real life. A CPU failed in a running system,
  80. becoming unresponsive, but not causing an immediate crash.
  81. This resulted in a series of RCU CPU stall warnings, eventually
  82. leading the realization that the CPU had failed.
  83. The RCU, RCU-sched, and RCU-tasks implementations have CPU stall warning.
  84. Note that SRCU does -not- have CPU stall warnings. Please note that
  85. RCU only detects CPU stalls when there is a grace period in progress.
  86. No grace period, no CPU stall warnings.
  87. To diagnose the cause of the stall, inspect the stack traces.
  88. The offending function will usually be near the top of the stack.
  89. If you have a series of stall warnings from a single extended stall,
  90. comparing the stack traces can often help determine where the stall
  91. is occurring, which will usually be in the function nearest the top of
  92. that portion of the stack which remains the same from trace to trace.
  93. If you can reliably trigger the stall, ftrace can be quite helpful.
  94. RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE
  95. and with RCU's event tracing. For information on RCU's event tracing,
  96. see include/trace/events/rcu.h.
  97. Fine-Tuning the RCU CPU Stall Detector
  98. ======================================
  99. The rcuupdate.rcu_cpu_stall_suppress module parameter disables RCU's
  100. CPU stall detector, which detects conditions that unduly delay RCU grace
  101. periods. This module parameter enables CPU stall detection by default,
  102. but may be overridden via boot-time parameter or at runtime via sysfs.
  103. The stall detector's idea of what constitutes "unduly delayed" is
  104. controlled by a set of kernel configuration variables and cpp macros:
  105. CONFIG_RCU_CPU_STALL_TIMEOUT
  106. ----------------------------
  107. This kernel configuration parameter defines the period of time
  108. that RCU will wait from the beginning of a grace period until it
  109. issues an RCU CPU stall warning. This time period is normally
  110. 21 seconds.
  111. This configuration parameter may be changed at runtime via the
  112. /sys/module/rcupdate/parameters/rcu_cpu_stall_timeout, however
  113. this parameter is checked only at the beginning of a cycle.
  114. So if you are 10 seconds into a 40-second stall, setting this
  115. sysfs parameter to (say) five will shorten the timeout for the
  116. -next- stall, or the following warning for the current stall
  117. (assuming the stall lasts long enough). It will not affect the
  118. timing of the next warning for the current stall.
  119. Stall-warning messages may be enabled and disabled completely via
  120. /sys/module/rcupdate/parameters/rcu_cpu_stall_suppress.
  121. RCU_STALL_DELAY_DELTA
  122. ---------------------
  123. Although the lockdep facility is extremely useful, it does add
  124. some overhead. Therefore, under CONFIG_PROVE_RCU, the
  125. RCU_STALL_DELAY_DELTA macro allows five extra seconds before
  126. giving an RCU CPU stall warning message. (This is a cpp
  127. macro, not a kernel configuration parameter.)
  128. RCU_STALL_RAT_DELAY
  129. -------------------
  130. The CPU stall detector tries to make the offending CPU print its
  131. own warnings, as this often gives better-quality stack traces.
  132. However, if the offending CPU does not detect its own stall in
  133. the number of jiffies specified by RCU_STALL_RAT_DELAY, then
  134. some other CPU will complain. This delay is normally set to
  135. two jiffies. (This is a cpp macro, not a kernel configuration
  136. parameter.)
  137. rcupdate.rcu_task_stall_timeout
  138. -------------------------------
  139. This boot/sysfs parameter controls the RCU-tasks stall warning
  140. interval. A value of zero or less suppresses RCU-tasks stall
  141. warnings. A positive value sets the stall-warning interval
  142. in seconds. An RCU-tasks stall warning starts with the line:
  143. INFO: rcu_tasks detected stalls on tasks:
  144. And continues with the output of sched_show_task() for each
  145. task stalling the current RCU-tasks grace period.
  146. Interpreting RCU's CPU Stall-Detector "Splats"
  147. ==============================================
  148. For non-RCU-tasks flavors of RCU, when a CPU detects that it is stalling,
  149. it will print a message similar to the following::
  150. INFO: rcu_sched detected stalls on CPUs/tasks:
  151. 2-...: (3 GPs behind) idle=06c/0/0 softirq=1453/1455 fqs=0
  152. 16-...: (0 ticks this GP) idle=81c/0/0 softirq=764/764 fqs=0
  153. (detected by 32, t=2603 jiffies, g=7075, q=625)
  154. This message indicates that CPU 32 detected that CPUs 2 and 16 were both
  155. causing stalls, and that the stall was affecting RCU-sched. This message
  156. will normally be followed by stack dumps for each CPU. Please note that
  157. PREEMPT_RCU builds can be stalled by tasks as well as by CPUs, and that
  158. the tasks will be indicated by PID, for example, "P3421". It is even
  159. possible for an rcu_state stall to be caused by both CPUs -and- tasks,
  160. in which case the offending CPUs and tasks will all be called out in the list.
  161. CPU 2's "(3 GPs behind)" indicates that this CPU has not interacted with
  162. the RCU core for the past three grace periods. In contrast, CPU 16's "(0
  163. ticks this GP)" indicates that this CPU has not taken any scheduling-clock
  164. interrupts during the current stalled grace period.
  165. The "idle=" portion of the message prints the dyntick-idle state.
  166. The hex number before the first "/" is the low-order 12 bits of the
  167. dynticks counter, which will have an even-numbered value if the CPU
  168. is in dyntick-idle mode and an odd-numbered value otherwise. The hex
  169. number between the two "/"s is the value of the nesting, which will be
  170. a small non-negative number if in the idle loop (as shown above) and a
  171. very large positive number otherwise.
  172. The "softirq=" portion of the message tracks the number of RCU softirq
  173. handlers that the stalled CPU has executed. The number before the "/"
  174. is the number that had executed since boot at the time that this CPU
  175. last noted the beginning of a grace period, which might be the current
  176. (stalled) grace period, or it might be some earlier grace period (for
  177. example, if the CPU might have been in dyntick-idle mode for an extended
  178. time period. The number after the "/" is the number that have executed
  179. since boot until the current time. If this latter number stays constant
  180. across repeated stall-warning messages, it is possible that RCU's softirq
  181. handlers are no longer able to execute on this CPU. This can happen if
  182. the stalled CPU is spinning with interrupts are disabled, or, in -rt
  183. kernels, if a high-priority process is starving RCU's softirq handler.
  184. The "fqs=" shows the number of force-quiescent-state idle/offline
  185. detection passes that the grace-period kthread has made across this
  186. CPU since the last time that this CPU noted the beginning of a grace
  187. period.
  188. The "detected by" line indicates which CPU detected the stall (in this
  189. case, CPU 32), how many jiffies have elapsed since the start of the grace
  190. period (in this case 2603), the grace-period sequence number (7075), and
  191. an estimate of the total number of RCU callbacks queued across all CPUs
  192. (625 in this case).
  193. In kernels with CONFIG_RCU_FAST_NO_HZ, more information is printed
  194. for each CPU::
  195. 0: (64628 ticks this GP) idle=dd5/3fffffffffffffff/0 softirq=82/543 last_accelerate: a345/d342 dyntick_enabled: 1
  196. The "last_accelerate:" prints the low-order 16 bits (in hex) of the
  197. jiffies counter when this CPU last invoked rcu_try_advance_all_cbs()
  198. from rcu_needs_cpu() or last invoked rcu_accelerate_cbs() from
  199. rcu_prepare_for_idle(). "dyntick_enabled: 1" indicates that dyntick-idle
  200. processing is enabled.
  201. If the grace period ends just as the stall warning starts printing,
  202. there will be a spurious stall-warning message, which will include
  203. the following::
  204. INFO: Stall ended before state dump start
  205. This is rare, but does happen from time to time in real life. It is also
  206. possible for a zero-jiffy stall to be flagged in this case, depending
  207. on how the stall warning and the grace-period initialization happen to
  208. interact. Please note that it is not possible to entirely eliminate this
  209. sort of false positive without resorting to things like stop_machine(),
  210. which is overkill for this sort of problem.
  211. If all CPUs and tasks have passed through quiescent states, but the
  212. grace period has nevertheless failed to end, the stall-warning splat
  213. will include something like the following::
  214. All QSes seen, last rcu_preempt kthread activity 23807 (4297905177-4297881370), jiffies_till_next_fqs=3, root ->qsmask 0x0
  215. The "23807" indicates that it has been more than 23 thousand jiffies
  216. since the grace-period kthread ran. The "jiffies_till_next_fqs"
  217. indicates how frequently that kthread should run, giving the number
  218. of jiffies between force-quiescent-state scans, in this case three,
  219. which is way less than 23807. Finally, the root rcu_node structure's
  220. ->qsmask field is printed, which will normally be zero.
  221. If the relevant grace-period kthread has been unable to run prior to
  222. the stall warning, as was the case in the "All QSes seen" line above,
  223. the following additional line is printed::
  224. kthread starved for 23807 jiffies! g7075 f0x0 RCU_GP_WAIT_FQS(3) ->state=0x1 ->cpu=5
  225. Starving the grace-period kthreads of CPU time can of course result
  226. in RCU CPU stall warnings even when all CPUs and tasks have passed
  227. through the required quiescent states. The "g" number shows the current
  228. grace-period sequence number, the "f" precedes the ->gp_flags command
  229. to the grace-period kthread, the "RCU_GP_WAIT_FQS" indicates that the
  230. kthread is waiting for a short timeout, the "state" precedes value of the
  231. task_struct ->state field, and the "cpu" indicates that the grace-period
  232. kthread last ran on CPU 5.
  233. Multiple Warnings From One Stall
  234. ================================
  235. If a stall lasts long enough, multiple stall-warning messages will be
  236. printed for it. The second and subsequent messages are printed at
  237. longer intervals, so that the time between (say) the first and second
  238. message will be about three times the interval between the beginning
  239. of the stall and the first message.
  240. Stall Warnings for Expedited Grace Periods
  241. ==========================================
  242. If an expedited grace period detects a stall, it will place a message
  243. like the following in dmesg::
  244. INFO: rcu_sched detected expedited stalls on CPUs/tasks: { 7-... } 21119 jiffies s: 73 root: 0x2/.
  245. This indicates that CPU 7 has failed to respond to a reschedule IPI.
  246. The three periods (".") following the CPU number indicate that the CPU
  247. is online (otherwise the first period would instead have been "O"),
  248. that the CPU was online at the beginning of the expedited grace period
  249. (otherwise the second period would have instead been "o"), and that
  250. the CPU has been online at least once since boot (otherwise, the third
  251. period would instead have been "N"). The number before the "jiffies"
  252. indicates that the expedited grace period has been going on for 21,119
  253. jiffies. The number following the "s:" indicates that the expedited
  254. grace-period sequence counter is 73. The fact that this last value is
  255. odd indicates that an expedited grace period is in flight. The number
  256. following "root:" is a bitmask that indicates which children of the root
  257. rcu_node structure correspond to CPUs and/or tasks that are blocking the
  258. current expedited grace period. If the tree had more than one level,
  259. additional hex numbers would be printed for the states of the other
  260. rcu_node structures in the tree.
  261. As with normal grace periods, PREEMPT_RCU builds can be stalled by
  262. tasks as well as by CPUs, and that the tasks will be indicated by PID,
  263. for example, "P3421".
  264. It is entirely possible to see stall warnings from normal and from
  265. expedited grace periods at about the same time during the same run.