1 Using RCU's CPU Stall Detector
3 This document first discusses what sorts of issues RCU's CPU stall
4 detector can locate, and then discusses kernel parameters and Kconfig
5 options that can be used to fine-tune the detector's operation. Finally,
6 this document explains the stall detector's "splat" format.
9 What Causes RCU CPU Stall Warnings?
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
15 o A CPU looping in an RCU read-side critical section.
17 o A CPU looping with interrupts disabled.
19 o A CPU looping with preemption disabled. This condition can
20 result in RCU-sched stalls and, if ksoftirqd is in use, RCU-bh
23 o A CPU looping with bottom halves disabled. This condition can
24 result in RCU-sched and RCU-bh stalls.
26 o For !CONFIG_PREEMPT kernels, a CPU looping anywhere in the kernel
27 without invoking schedule(). If the looping in the kernel is
28 really expected and desirable behavior, you might need to add
29 some calls to cond_resched().
31 o Booting Linux using a console connection that is too slow to
32 keep up with the boot-time console-message rate. For example,
33 a 115Kbaud serial console can be -way- too slow to keep up
34 with boot-time message rates, and will frequently result in
35 RCU CPU stall warning messages. Especially if you have added
38 o Anything that prevents RCU's grace-period kthreads from running.
39 This can result in the "All QSes seen" console-log message.
40 This message will include information on when the kthread last
41 ran and how often it should be expected to run. It can also
42 result in the "rcu_.*kthread starved for" console-log message,
43 which will include additional debugging information.
45 o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might
46 happen to preempt a low-priority task in the middle of an RCU
47 read-side critical section. This is especially damaging if
48 that low-priority task is not permitted to run on any other CPU,
49 in which case the next RCU grace period can never complete, which
50 will eventually cause the system to run out of memory and hang.
51 While the system is in the process of running itself out of
52 memory, you might see stall-warning messages.
54 o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that
55 is running at a higher priority than the RCU softirq threads.
56 This will prevent RCU callbacks from ever being invoked,
57 and in a CONFIG_PREEMPT_RCU kernel will further prevent
58 RCU grace periods from ever completing. Either way, the
59 system will eventually run out of memory and hang. In the
60 CONFIG_PREEMPT_RCU case, you might see stall-warning
63 o A periodic interrupt whose handler takes longer than the time
64 interval between successive pairs of interrupts. This can
65 prevent RCU's kthreads and softirq handlers from running.
66 Note that certain high-overhead debugging options, for example
67 the function_graph tracer, can result in interrupt handler taking
68 considerably longer than normal, which can in turn result in
69 RCU CPU stall warnings.
71 o Testing a workload on a fast system, tuning the stall-warning
72 timeout down to just barely avoid RCU CPU stall warnings, and then
73 running the same workload with the same stall-warning timeout on a
74 slow system. Note that thermal throttling and on-demand governors
75 can cause a single system to be sometimes fast and sometimes slow!
77 o A hardware or software issue shuts off the scheduler-clock
78 interrupt on a CPU that is not in dyntick-idle mode. This
79 problem really has happened, and seems to be most likely to
80 result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels.
82 o A bug in the RCU implementation.
84 o A hardware failure. This is quite unlikely, but has occurred
85 at least once in real life. A CPU failed in a running system,
86 becoming unresponsive, but not causing an immediate crash.
87 This resulted in a series of RCU CPU stall warnings, eventually
88 leading the realization that the CPU had failed.
90 The RCU, RCU-sched, RCU-bh, and RCU-tasks implementations have CPU stall
91 warning. Note that SRCU does -not- have CPU stall warnings. Please note
92 that RCU only detects CPU stalls when there is a grace period in progress.
93 No grace period, no CPU stall warnings.
95 To diagnose the cause of the stall, inspect the stack traces.
96 The offending function will usually be near the top of the stack.
97 If you have a series of stall warnings from a single extended stall,
98 comparing the stack traces can often help determine where the stall
99 is occurring, which will usually be in the function nearest the top of
100 that portion of the stack which remains the same from trace to trace.
101 If you can reliably trigger the stall, ftrace can be quite helpful.
103 RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE
104 and with RCU's event tracing. For information on RCU's event tracing,
105 see include/trace/events/rcu.h.
108 Fine-Tuning the RCU CPU Stall Detector
110 The rcuupdate.rcu_cpu_stall_suppress module parameter disables RCU's
111 CPU stall detector, which detects conditions that unduly delay RCU grace
112 periods. This module parameter enables CPU stall detection by default,
113 but may be overridden via boot-time parameter or at runtime via sysfs.
114 The stall detector's idea of what constitutes "unduly delayed" is
115 controlled by a set of kernel configuration variables and cpp macros:
117 CONFIG_RCU_CPU_STALL_TIMEOUT
119 This kernel configuration parameter defines the period of time
120 that RCU will wait from the beginning of a grace period until it
121 issues an RCU CPU stall warning. This time period is normally
124 This configuration parameter may be changed at runtime via the
125 /sys/module/rcupdate/parameters/rcu_cpu_stall_timeout, however
126 this parameter is checked only at the beginning of a cycle.
127 So if you are 10 seconds into a 40-second stall, setting this
128 sysfs parameter to (say) five will shorten the timeout for the
129 -next- stall, or the following warning for the current stall
130 (assuming the stall lasts long enough). It will not affect the
131 timing of the next warning for the current stall.
133 Stall-warning messages may be enabled and disabled completely via
134 /sys/module/rcupdate/parameters/rcu_cpu_stall_suppress.
136 RCU_STALL_DELAY_DELTA
138 Although the lockdep facility is extremely useful, it does add
139 some overhead. Therefore, under CONFIG_PROVE_RCU, the
140 RCU_STALL_DELAY_DELTA macro allows five extra seconds before
141 giving an RCU CPU stall warning message. (This is a cpp
142 macro, not a kernel configuration parameter.)
146 The CPU stall detector tries to make the offending CPU print its
147 own warnings, as this often gives better-quality stack traces.
148 However, if the offending CPU does not detect its own stall in
149 the number of jiffies specified by RCU_STALL_RAT_DELAY, then
150 some other CPU will complain. This delay is normally set to
151 two jiffies. (This is a cpp macro, not a kernel configuration
154 rcupdate.rcu_task_stall_timeout
156 This boot/sysfs parameter controls the RCU-tasks stall warning
157 interval. A value of zero or less suppresses RCU-tasks stall
158 warnings. A positive value sets the stall-warning interval
159 in jiffies. An RCU-tasks stall warning starts with the line:
161 INFO: rcu_tasks detected stalls on tasks:
163 And continues with the output of sched_show_task() for each
164 task stalling the current RCU-tasks grace period.
167 Interpreting RCU's CPU Stall-Detector "Splats"
169 For non-RCU-tasks flavors of RCU, when a CPU detects that it is stalling,
170 it will print a message similar to the following:
172 INFO: rcu_sched detected stalls on CPUs/tasks:
173 2-...: (3 GPs behind) idle=06c/0/0 softirq=1453/1455 fqs=0
174 16-...: (0 ticks this GP) idle=81c/0/0 softirq=764/764 fqs=0
175 (detected by 32, t=2603 jiffies, g=7073, c=7072, q=625)
177 This message indicates that CPU 32 detected that CPUs 2 and 16 were both
178 causing stalls, and that the stall was affecting RCU-sched. This message
179 will normally be followed by stack dumps for each CPU. Please note that
180 PREEMPT_RCU builds can be stalled by tasks as well as by CPUs, and that
181 the tasks will be indicated by PID, for example, "P3421". It is even
182 possible for a rcu_preempt_state stall to be caused by both CPUs -and-
183 tasks, in which case the offending CPUs and tasks will all be called
186 CPU 2's "(3 GPs behind)" indicates that this CPU has not interacted with
187 the RCU core for the past three grace periods. In contrast, CPU 16's "(0
188 ticks this GP)" indicates that this CPU has not taken any scheduling-clock
189 interrupts during the current stalled grace period.
191 The "idle=" portion of the message prints the dyntick-idle state.
192 The hex number before the first "/" is the low-order 12 bits of the
193 dynticks counter, which will have an even-numbered value if the CPU
194 is in dyntick-idle mode and an odd-numbered value otherwise. The hex
195 number between the two "/"s is the value of the nesting, which will be
196 a small non-negative number if in the idle loop (as shown above) and a
197 very large positive number otherwise.
199 The "softirq=" portion of the message tracks the number of RCU softirq
200 handlers that the stalled CPU has executed. The number before the "/"
201 is the number that had executed since boot at the time that this CPU
202 last noted the beginning of a grace period, which might be the current
203 (stalled) grace period, or it might be some earlier grace period (for
204 example, if the CPU might have been in dyntick-idle mode for an extended
205 time period. The number after the "/" is the number that have executed
206 since boot until the current time. If this latter number stays constant
207 across repeated stall-warning messages, it is possible that RCU's softirq
208 handlers are no longer able to execute on this CPU. This can happen if
209 the stalled CPU is spinning with interrupts are disabled, or, in -rt
210 kernels, if a high-priority process is starving RCU's softirq handler.
212 The "fps=" shows the number of force-quiescent-state idle/offline
213 detection passes that the grace-period kthread has made across this
214 CPU since the last time that this CPU noted the beginning of a grace
217 The "detected by" line indicates which CPU detected the stall (in this
218 case, CPU 32), how many jiffies have elapsed since the start of the
219 grace period (in this case 2603), the number of the last grace period
220 to start and to complete (7073 and 7072, respectively), and an estimate
221 of the total number of RCU callbacks queued across all CPUs (625 in
224 In kernels with CONFIG_RCU_FAST_NO_HZ, more information is printed
227 0: (64628 ticks this GP) idle=dd5/3fffffffffffffff/0 softirq=82/543 last_accelerate: a345/d342 nonlazy_posted: 25 .D
229 The "last_accelerate:" prints the low-order 16 bits (in hex) of the
230 jiffies counter when this CPU last invoked rcu_try_advance_all_cbs()
231 from rcu_needs_cpu() or last invoked rcu_accelerate_cbs() from
232 rcu_prepare_for_idle(). The "nonlazy_posted:" prints the number
233 of non-lazy callbacks posted since the last call to rcu_needs_cpu().
234 Finally, an "L" indicates that there are currently no non-lazy callbacks
235 ("." is printed otherwise, as shown above) and "D" indicates that
236 dyntick-idle processing is enabled ("." is printed otherwise, for example,
237 if disabled via the "nohz=" kernel boot parameter).
239 If the grace period ends just as the stall warning starts printing,
240 there will be a spurious stall-warning message, which will include
243 INFO: Stall ended before state dump start
245 This is rare, but does happen from time to time in real life. It is also
246 possible for a zero-jiffy stall to be flagged in this case, depending
247 on how the stall warning and the grace-period initialization happen to
248 interact. Please note that it is not possible to entirely eliminate this
249 sort of false positive without resorting to things like stop_machine(),
250 which is overkill for this sort of problem.
252 If all CPUs and tasks have passed through quiescent states, but the
253 grace period has nevertheless failed to end, the stall-warning splat
254 will include something like the following:
256 All QSes seen, last rcu_preempt kthread activity 23807 (4297905177-4297881370), jiffies_till_next_fqs=3, root ->qsmask 0x0
258 The "23807" indicates that it has been more than 23 thousand jiffies
259 since the grace-period kthread ran. The "jiffies_till_next_fqs"
260 indicates how frequently that kthread should run, giving the number
261 of jiffies between force-quiescent-state scans, in this case three,
262 which is way less than 23807. Finally, the root rcu_node structure's
263 ->qsmask field is printed, which will normally be zero.
265 If the relevant grace-period kthread has been unable to run prior to
266 the stall warning, as was the case in the "All QSes seen" line above,
267 the following additional line is printed:
269 kthread starved for 23807 jiffies! g7073 c7072 f0x0 RCU_GP_WAIT_FQS(3) ->state=0x1
271 Starving the grace-period kthreads of CPU time can of course result
272 in RCU CPU stall warnings even when all CPUs and tasks have passed
273 through the required quiescent states. The "g" and "c" numbers flag the
274 number of the last grace period started and completed, respectively,
275 the "f" precedes the ->gp_flags command to the grace-period kthread,
276 the "RCU_GP_WAIT_FQS" indicates that the kthread is waiting for a short
277 timeout, and the "state" precedes value of the task_struct ->state field.
280 Multiple Warnings From One Stall
282 If a stall lasts long enough, multiple stall-warning messages will be
283 printed for it. The second and subsequent messages are printed at
284 longer intervals, so that the time between (say) the first and second
285 message will be about three times the interval between the beginning
286 of the stall and the first message.
289 Stall Warnings for Expedited Grace Periods
291 If an expedited grace period detects a stall, it will place a message
292 like the following in dmesg:
294 INFO: rcu_sched detected expedited stalls on CPUs/tasks: { 7-... } 21119 jiffies s: 73 root: 0x2/.
296 This indicates that CPU 7 has failed to respond to a reschedule IPI.
297 The three periods (".") following the CPU number indicate that the CPU
298 is online (otherwise the first period would instead have been "O"),
299 that the CPU was online at the beginning of the expedited grace period
300 (otherwise the second period would have instead been "o"), and that
301 the CPU has been online at least once since boot (otherwise, the third
302 period would instead have been "N"). The number before the "jiffies"
303 indicates that the expedited grace period has been going on for 21,119
304 jiffies. The number following the "s:" indicates that the expedited
305 grace-period sequence counter is 73. The fact that this last value is
306 odd indicates that an expedited grace period is in flight. The number
307 following "root:" is a bitmask that indicates which children of the root
308 rcu_node structure correspond to CPUs and/or tasks that are blocking the
309 current expedited grace period. If the tree had more than one level,
310 additional hex numbers would be printed for the states of the other
311 rcu_node structures in the tree.
313 As with normal grace periods, PREEMPT_RCU builds can be stalled by
314 tasks as well as by CPUs, and that the tasks will be indicated by PID,
315 for example, "P3421".
317 It is entirely possible to see stall warnings from normal and from
318 expedited grace periods at about the same time during the same run.