1 Using RCU's CPU Stall Detector
3 The rcu_cpu_stall_suppress module parameter enables RCU's CPU stall
4 detector, which detects conditions that unduly delay RCU grace periods.
5 This module parameter enables CPU stall detection by default, but
6 may be overridden via boot-time parameter or at runtime via sysfs.
7 The stall detector's idea of what constitutes "unduly delayed" is
8 controlled by a set of kernel configuration variables and cpp macros:
10 CONFIG_RCU_CPU_STALL_TIMEOUT
12 This kernel configuration parameter defines the period of time
13 that RCU will wait from the beginning of a grace period until it
14 issues an RCU CPU stall warning. This time period is normally
17 This configuration parameter may be changed at runtime via the
18 /sys/module/rcupdate/parameters/rcu_cpu_stall_timeout, however
19 this parameter is checked only at the beginning of a cycle.
20 So if you are 10 seconds into a 40-second stall, setting this
21 sysfs parameter to (say) five will shorten the timeout for the
22 -next- stall, or the following warning for the current stall
23 (assuming the stall lasts long enough). It will not affect the
24 timing of the next warning for the current stall.
26 Stall-warning messages may be enabled and disabled completely via
27 /sys/module/rcupdate/parameters/rcu_cpu_stall_suppress.
31 Although the lockdep facility is extremely useful, it does add
32 some overhead. Therefore, under CONFIG_PROVE_RCU, the
33 RCU_STALL_DELAY_DELTA macro allows five extra seconds before
34 giving an RCU CPU stall warning message. (This is a cpp
35 macro, not a kernel configuration parameter.)
39 The CPU stall detector tries to make the offending CPU print its
40 own warnings, as this often gives better-quality stack traces.
41 However, if the offending CPU does not detect its own stall in
42 the number of jiffies specified by RCU_STALL_RAT_DELAY, then
43 some other CPU will complain. This delay is normally set to
44 two jiffies. (This is a cpp macro, not a kernel configuration
47 rcupdate.rcu_task_stall_timeout
49 This boot/sysfs parameter controls the RCU-tasks stall warning
50 interval. A value of zero or less suppresses RCU-tasks stall
51 warnings. A positive value sets the stall-warning interval
52 in jiffies. An RCU-tasks stall warning starts wtih the line:
54 INFO: rcu_tasks detected stalls on tasks:
56 And continues with the output of sched_show_task() for each
57 task stalling the current RCU-tasks grace period.
59 For non-RCU-tasks flavors of RCU, when a CPU detects that it is stalling,
60 it will print a message similar to the following:
62 INFO: rcu_sched_state detected stall on CPU 5 (t=2500 jiffies)
64 This message indicates that CPU 5 detected that it was causing a stall,
65 and that the stall was affecting RCU-sched. This message will normally be
66 followed by a stack dump of the offending CPU. On TREE_RCU kernel builds,
67 RCU and RCU-sched are implemented by the same underlying mechanism,
68 while on PREEMPT_RCU kernel builds, RCU is instead implemented
71 On the other hand, if the offending CPU fails to print out a stall-warning
72 message quickly enough, some other CPU will print a message similar to
75 INFO: rcu_bh_state detected stalls on CPUs/tasks: { 3 5 } (detected by 2, 2502 jiffies)
77 This message indicates that CPU 2 detected that CPUs 3 and 5 were both
78 causing stalls, and that the stall was affecting RCU-bh. This message
79 will normally be followed by stack dumps for each CPU. Please note that
80 PREEMPT_RCU builds can be stalled by tasks as well as by CPUs,
81 and that the tasks will be indicated by PID, for example, "P3421".
82 It is even possible for a rcu_preempt_state stall to be caused by both
83 CPUs -and- tasks, in which case the offending CPUs and tasks will all
84 be called out in the list.
86 Finally, if the grace period ends just as the stall warning starts
87 printing, there will be a spurious stall-warning message:
89 INFO: rcu_bh_state detected stalls on CPUs/tasks: { } (detected by 4, 2502 jiffies)
91 This is rare, but does happen from time to time in real life. It is also
92 possible for a zero-jiffy stall to be flagged in this case, depending
93 on how the stall warning and the grace-period initialization happen to
94 interact. Please note that it is not possible to entirely eliminate this
95 sort of false positive without resorting to things like stop_machine(),
96 which is overkill for this sort of problem.
98 Recent kernels will print a long form of the stall-warning message:
100 INFO: rcu_preempt detected stall on CPU
101 0: (63959 ticks this GP) idle=241/3fffffffffffffff/0 softirq=82/543
104 In kernels with CONFIG_RCU_FAST_NO_HZ, more information is printed:
106 INFO: rcu_preempt detected stall on CPU
107 0: (64628 ticks this GP) idle=dd5/3fffffffffffffff/0 softirq=82/543 last_accelerate: a345/d342 nonlazy_posted: 25 .D
110 The "(64628 ticks this GP)" indicates that this CPU has taken more
111 than 64,000 scheduling-clock interrupts during the current stalled
112 grace period. If the CPU was not yet aware of the current grace
113 period (for example, if it was offline), then this part of the message
114 indicates how many grace periods behind the CPU is.
116 The "idle=" portion of the message prints the dyntick-idle state.
117 The hex number before the first "/" is the low-order 12 bits of the
118 dynticks counter, which will have an even-numbered value if the CPU is
119 in dyntick-idle mode and an odd-numbered value otherwise. The hex
120 number between the two "/"s is the value of the nesting, which will
121 be a small positive number if in the idle loop and a very large positive
122 number (as shown above) otherwise.
124 The "softirq=" portion of the message tracks the number of RCU softirq
125 handlers that the stalled CPU has executed. The number before the "/"
126 is the number that had executed since boot at the time that this CPU
127 last noted the beginning of a grace period, which might be the current
128 (stalled) grace period, or it might be some earlier grace period (for
129 example, if the CPU might have been in dyntick-idle mode for an extended
130 time period. The number after the "/" is the number that have executed
131 since boot until the current time. If this latter number stays constant
132 across repeated stall-warning messages, it is possible that RCU's softirq
133 handlers are no longer able to execute on this CPU. This can happen if
134 the stalled CPU is spinning with interrupts are disabled, or, in -rt
135 kernels, if a high-priority process is starving RCU's softirq handler.
137 For CONFIG_RCU_FAST_NO_HZ kernels, the "last_accelerate:" prints the
138 low-order 16 bits (in hex) of the jiffies counter when this CPU last
139 invoked rcu_try_advance_all_cbs() from rcu_needs_cpu() or last invoked
140 rcu_accelerate_cbs() from rcu_prepare_for_idle(). The "nonlazy_posted:"
141 prints the number of non-lazy callbacks posted since the last call to
142 rcu_needs_cpu(). Finally, an "L" indicates that there are currently
143 no non-lazy callbacks ("." is printed otherwise, as shown above) and
144 "D" indicates that dyntick-idle processing is enabled ("." is printed
145 otherwise, for example, if disabled via the "nohz=" kernel boot parameter).
147 If the relevant grace-period kthread has been unable to run prior to
148 the stall warning, the following additional line is printed:
150 rcu_preempt kthread starved for 2023 jiffies!
152 Starving the grace-period kthreads of CPU time can of course result in
153 RCU CPU stall warnings even when all CPUs and tasks have passed through
154 the required quiescent states.
157 Multiple Warnings From One Stall
159 If a stall lasts long enough, multiple stall-warning messages will be
160 printed for it. The second and subsequent messages are printed at
161 longer intervals, so that the time between (say) the first and second
162 message will be about three times the interval between the beginning
163 of the stall and the first message.
166 Stall Warnings for Expedited Grace Periods
168 If an expedited grace period detects a stall, it will place a message
169 like the following in dmesg:
171 INFO: rcu_sched detected expedited stalls on CPUs: { 1 2 6 } 26009 jiffies s: 1043
173 This indicates that CPUs 1, 2, and 6 have failed to respond to a
174 reschedule IPI, that the expedited grace period has been going on for
175 26,009 jiffies, and that the expedited grace-period sequence counter is
176 1043. The fact that this last value is odd indicates that an expedited
177 grace period is in flight.
179 It is entirely possible to see stall warnings from normal and from
180 expedited grace periods at about the same time from the same run.
183 What Causes RCU CPU Stall Warnings?
185 So your kernel printed an RCU CPU stall warning. The next question is
186 "What caused it?" The following problems can result in RCU CPU stall
189 o A CPU looping in an RCU read-side critical section.
191 o A CPU looping with interrupts disabled. This condition can
192 result in RCU-sched and RCU-bh stalls.
194 o A CPU looping with preemption disabled. This condition can
195 result in RCU-sched stalls and, if ksoftirqd is in use, RCU-bh
198 o A CPU looping with bottom halves disabled. This condition can
199 result in RCU-sched and RCU-bh stalls.
201 o For !CONFIG_PREEMPT kernels, a CPU looping anywhere in the
202 kernel without invoking schedule(). Note that cond_resched()
203 does not necessarily prevent RCU CPU stall warnings. Therefore,
204 if the looping in the kernel is really expected and desirable
205 behavior, you might need to replace some of the cond_resched()
206 calls with calls to cond_resched_rcu_qs().
208 o Anything that prevents RCU's grace-period kthreads from running.
209 This can result in the "All QSes seen" console-log message.
210 This message will include information on when the kthread last
211 ran and how often it should be expected to run.
213 o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might
214 happen to preempt a low-priority task in the middle of an RCU
215 read-side critical section. This is especially damaging if
216 that low-priority task is not permitted to run on any other CPU,
217 in which case the next RCU grace period can never complete, which
218 will eventually cause the system to run out of memory and hang.
219 While the system is in the process of running itself out of
220 memory, you might see stall-warning messages.
222 o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that
223 is running at a higher priority than the RCU softirq threads.
224 This will prevent RCU callbacks from ever being invoked,
225 and in a CONFIG_PREEMPT_RCU kernel will further prevent
226 RCU grace periods from ever completing. Either way, the
227 system will eventually run out of memory and hang. In the
228 CONFIG_PREEMPT_RCU case, you might see stall-warning
231 o A hardware or software issue shuts off the scheduler-clock
232 interrupt on a CPU that is not in dyntick-idle mode. This
233 problem really has happened, and seems to be most likely to
234 result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels.
236 o A bug in the RCU implementation.
238 o A hardware failure. This is quite unlikely, but has occurred
239 at least once in real life. A CPU failed in a running system,
240 becoming unresponsive, but not causing an immediate crash.
241 This resulted in a series of RCU CPU stall warnings, eventually
242 leading the realization that the CPU had failed.
244 The RCU, RCU-sched, RCU-bh, and RCU-tasks implementations have CPU stall
245 warning. Note that SRCU does -not- have CPU stall warnings. Please note
246 that RCU only detects CPU stalls when there is a grace period in progress.
247 No grace period, no CPU stall warnings.
249 To diagnose the cause of the stall, inspect the stack traces.
250 The offending function will usually be near the top of the stack.
251 If you have a series of stall warnings from a single extended stall,
252 comparing the stack traces can often help determine where the stall
253 is occurring, which will usually be in the function nearest the top of
254 that portion of the stack which remains the same from trace to trace.
255 If you can reliably trigger the stall, ftrace can be quite helpful.
257 RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE
258 and with RCU's event tracing. For information on RCU's event tracing,
259 see include/trace/events/rcu.h.