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.
29 CONFIG_RCU_CPU_STALL_INFO
31 This kernel configuration parameter causes the stall warning to
32 print out additional per-CPU diagnostic information, including
33 information on scheduling-clock ticks and RCU's idle-CPU tracking.
37 Although the lockdep facility is extremely useful, it does add
38 some overhead. Therefore, under CONFIG_PROVE_RCU, the
39 RCU_STALL_DELAY_DELTA macro allows five extra seconds before
40 giving an RCU CPU stall warning message. (This is a cpp
41 macro, not a kernel configuration parameter.)
45 The CPU stall detector tries to make the offending CPU print its
46 own warnings, as this often gives better-quality stack traces.
47 However, if the offending CPU does not detect its own stall in
48 the number of jiffies specified by RCU_STALL_RAT_DELAY, then
49 some other CPU will complain. This delay is normally set to
50 two jiffies. (This is a cpp macro, not a kernel configuration
53 rcupdate.rcu_task_stall_timeout
55 This boot/sysfs parameter controls the RCU-tasks stall warning
56 interval. A value of zero or less suppresses RCU-tasks stall
57 warnings. A positive value sets the stall-warning interval
58 in jiffies. An RCU-tasks stall warning starts wtih the line:
60 INFO: rcu_tasks detected stalls on tasks:
62 And continues with the output of sched_show_task() for each
63 task stalling the current RCU-tasks grace period.
65 For non-RCU-tasks flavors of RCU, when a CPU detects that it is stalling,
66 it will print a message similar to the following:
68 INFO: rcu_sched_state detected stall on CPU 5 (t=2500 jiffies)
70 This message indicates that CPU 5 detected that it was causing a stall,
71 and that the stall was affecting RCU-sched. This message will normally be
72 followed by a stack dump of the offending CPU. On TREE_RCU kernel builds,
73 RCU and RCU-sched are implemented by the same underlying mechanism,
74 while on PREEMPT_RCU kernel builds, RCU is instead implemented
77 On the other hand, if the offending CPU fails to print out a stall-warning
78 message quickly enough, some other CPU will print a message similar to
81 INFO: rcu_bh_state detected stalls on CPUs/tasks: { 3 5 } (detected by 2, 2502 jiffies)
83 This message indicates that CPU 2 detected that CPUs 3 and 5 were both
84 causing stalls, and that the stall was affecting RCU-bh. This message
85 will normally be followed by stack dumps for each CPU. Please note that
86 PREEMPT_RCU builds can be stalled by tasks as well as by CPUs,
87 and that the tasks will be indicated by PID, for example, "P3421".
88 It is even possible for a rcu_preempt_state stall to be caused by both
89 CPUs -and- tasks, in which case the offending CPUs and tasks will all
90 be called out in the list.
92 Finally, if the grace period ends just as the stall warning starts
93 printing, there will be a spurious stall-warning message:
95 INFO: rcu_bh_state detected stalls on CPUs/tasks: { } (detected by 4, 2502 jiffies)
97 This is rare, but does happen from time to time in real life. It is also
98 possible for a zero-jiffy stall to be flagged in this case, depending
99 on how the stall warning and the grace-period initialization happen to
100 interact. Please note that it is not possible to entirely eliminate this
101 sort of false positive without resorting to things like stop_machine(),
102 which is overkill for this sort of problem.
104 If the CONFIG_RCU_CPU_STALL_INFO kernel configuration parameter is set,
105 more information is printed with the stall-warning message, for example:
107 INFO: rcu_preempt detected stall on CPU
108 0: (63959 ticks this GP) idle=241/3fffffffffffffff/0 softirq=82/543
111 In kernels with CONFIG_RCU_FAST_NO_HZ, even more information is
114 INFO: rcu_preempt detected stall on CPU
115 0: (64628 ticks this GP) idle=dd5/3fffffffffffffff/0 softirq=82/543 last_accelerate: a345/d342 nonlazy_posted: 25 .D
118 The "(64628 ticks this GP)" indicates that this CPU has taken more
119 than 64,000 scheduling-clock interrupts during the current stalled
120 grace period. If the CPU was not yet aware of the current grace
121 period (for example, if it was offline), then this part of the message
122 indicates how many grace periods behind the CPU is.
124 The "idle=" portion of the message prints the dyntick-idle state.
125 The hex number before the first "/" is the low-order 12 bits of the
126 dynticks counter, which will have an even-numbered value if the CPU is
127 in dyntick-idle mode and an odd-numbered value otherwise. The hex
128 number between the two "/"s is the value of the nesting, which will
129 be a small positive number if in the idle loop and a very large positive
130 number (as shown above) otherwise.
132 The "softirq=" portion of the message tracks the number of RCU softirq
133 handlers that the stalled CPU has executed. The number before the "/"
134 is the number that had executed since boot at the time that this CPU
135 last noted the beginning of a grace period, which might be the current
136 (stalled) grace period, or it might be some earlier grace period (for
137 example, if the CPU might have been in dyntick-idle mode for an extended
138 time period. The number after the "/" is the number that have executed
139 since boot until the current time. If this latter number stays constant
140 across repeated stall-warning messages, it is possible that RCU's softirq
141 handlers are no longer able to execute on this CPU. This can happen if
142 the stalled CPU is spinning with interrupts are disabled, or, in -rt
143 kernels, if a high-priority process is starving RCU's softirq handler.
145 For CONFIG_RCU_FAST_NO_HZ kernels, the "last_accelerate:" prints the
146 low-order 16 bits (in hex) of the jiffies counter when this CPU last
147 invoked rcu_try_advance_all_cbs() from rcu_needs_cpu() or last invoked
148 rcu_accelerate_cbs() from rcu_prepare_for_idle(). The "nonlazy_posted:"
149 prints the number of non-lazy callbacks posted since the last call to
150 rcu_needs_cpu(). Finally, an "L" indicates that there are currently
151 no non-lazy callbacks ("." is printed otherwise, as shown above) and
152 "D" indicates that dyntick-idle processing is enabled ("." is printed
153 otherwise, for example, if disabled via the "nohz=" kernel boot parameter).
155 If the relevant grace-period kthread has been unable to run prior to
156 the stall warning, the following additional line is printed:
158 rcu_preempt kthread starved for 2023 jiffies!
160 Starving the grace-period kthreads of CPU time can of course result in
161 RCU CPU stall warnings even when all CPUs and tasks have passed through
162 the required quiescent states.
165 Multiple Warnings From One Stall
167 If a stall lasts long enough, multiple stall-warning messages will be
168 printed for it. The second and subsequent messages are printed at
169 longer intervals, so that the time between (say) the first and second
170 message will be about three times the interval between the beginning
171 of the stall and the first message.
174 What Causes RCU CPU Stall Warnings?
176 So your kernel printed an RCU CPU stall warning. The next question is
177 "What caused it?" The following problems can result in RCU CPU stall
180 o A CPU looping in an RCU read-side critical section.
182 o A CPU looping with interrupts disabled. This condition can
183 result in RCU-sched and RCU-bh stalls.
185 o A CPU looping with preemption disabled. This condition can
186 result in RCU-sched stalls and, if ksoftirqd is in use, RCU-bh
189 o A CPU looping with bottom halves disabled. This condition can
190 result in RCU-sched and RCU-bh stalls.
192 o For !CONFIG_PREEMPT kernels, a CPU looping anywhere in the
193 kernel without invoking schedule(). Note that cond_resched()
194 does not necessarily prevent RCU CPU stall warnings. Therefore,
195 if the looping in the kernel is really expected and desirable
196 behavior, you might need to replace some of the cond_resched()
197 calls with calls to cond_resched_rcu_qs().
199 o Anything that prevents RCU's grace-period kthreads from running.
200 This can result in the "All QSes seen" console-log message.
201 This message will include information on when the kthread last
202 ran and how often it should be expected to run.
204 o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might
205 happen to preempt a low-priority task in the middle of an RCU
206 read-side critical section. This is especially damaging if
207 that low-priority task is not permitted to run on any other CPU,
208 in which case the next RCU grace period can never complete, which
209 will eventually cause the system to run out of memory and hang.
210 While the system is in the process of running itself out of
211 memory, you might see stall-warning messages.
213 o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that
214 is running at a higher priority than the RCU softirq threads.
215 This will prevent RCU callbacks from ever being invoked,
216 and in a CONFIG_PREEMPT_RCU kernel will further prevent
217 RCU grace periods from ever completing. Either way, the
218 system will eventually run out of memory and hang. In the
219 CONFIG_PREEMPT_RCU case, you might see stall-warning
222 o A hardware or software issue shuts off the scheduler-clock
223 interrupt on a CPU that is not in dyntick-idle mode. This
224 problem really has happened, and seems to be most likely to
225 result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels.
227 o A bug in the RCU implementation.
229 o A hardware failure. This is quite unlikely, but has occurred
230 at least once in real life. A CPU failed in a running system,
231 becoming unresponsive, but not causing an immediate crash.
232 This resulted in a series of RCU CPU stall warnings, eventually
233 leading the realization that the CPU had failed.
235 The RCU, RCU-sched, RCU-bh, and RCU-tasks implementations have CPU stall
236 warning. Note that SRCU does -not- have CPU stall warnings. Please note
237 that RCU only detects CPU stalls when there is a grace period in progress.
238 No grace period, no CPU stall warnings.
240 To diagnose the cause of the stall, inspect the stack traces.
241 The offending function will usually be near the top of the stack.
242 If you have a series of stall warnings from a single extended stall,
243 comparing the stack traces can often help determine where the stall
244 is occurring, which will usually be in the function nearest the top of
245 that portion of the stack which remains the same from trace to trace.
246 If you can reliably trigger the stall, ftrace can be quite helpful.
248 RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE
249 and with RCU's event tracing. For information on RCU's event tracing,
250 see include/trace/events/rcu.h.