blob: 8e9359de1d28b2e845d25425be72451b3f6be4f7 [file] [log] [blame]
Paul E. McKenney4c540052010-01-14 16:10:57 -08001Using RCU's CPU Stall Detector
2
Paul E. McKenneya00e0d712011-02-08 17:14:39 -08003The rcu_cpu_stall_suppress module parameter enables RCU's CPU stall
4detector, which detects conditions that unduly delay RCU grace periods.
5This module parameter enables CPU stall detection by default, but
6may be overridden via boot-time parameter or at runtime via sysfs.
7The stall detector's idea of what constitutes "unduly delayed" is
8controlled by a set of kernel configuration variables and cpp macros:
Paul E. McKenney4c540052010-01-14 16:10:57 -08009
Paul E. McKenneya00e0d712011-02-08 17:14:39 -080010CONFIG_RCU_CPU_STALL_TIMEOUT
Paul E. McKenney4c540052010-01-14 16:10:57 -080011
Paul E. McKenneya00e0d712011-02-08 17:14:39 -080012 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
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080015 sixty seconds.
Paul E. McKenney4c540052010-01-14 16:10:57 -080016
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080017 This configuration parameter may be changed at runtime via the
18 /sys/module/rcutree/parameters/rcu_cpu_stall_timeout, however
19 this parameter is checked only at the beginning of a cycle.
20 So if you are 30 seconds into a 70-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.
Paul E. McKenney4c540052010-01-14 16:10:57 -080025
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080026 Stall-warning messages may be enabled and disabled completely via
27 /sys/module/rcutree/parameters/rcu_cpu_stall_suppress.
28
29CONFIG_RCU_CPU_STALL_VERBOSE
30
31 This kernel configuration parameter causes the stall warning to
32 also dump the stacks of any tasks that are blocking the current
33 RCU-preempt grace period.
34
35RCU_CPU_STALL_INFO
36
37 This kernel configuration parameter causes the stall warning to
38 print out additional per-CPU diagnostic information, including
39 information on scheduling-clock ticks and RCU's idle-CPU tracking.
40
41RCU_STALL_DELAY_DELTA
42
43 Although the lockdep facility is extremely useful, it does add
44 some overhead. Therefore, under CONFIG_PROVE_RCU, the
45 RCU_STALL_DELAY_DELTA macro allows five extra seconds before
46 giving an RCU CPU stall warning message.
Paul E. McKenney4c540052010-01-14 16:10:57 -080047
48RCU_STALL_RAT_DELAY
49
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -070050 The CPU stall detector tries to make the offending CPU print its
51 own warnings, as this often gives better-quality stack traces.
52 However, if the offending CPU does not detect its own stall in
53 the number of jiffies specified by RCU_STALL_RAT_DELAY, then
54 some other CPU will complain. This delay is normally set to
55 two jiffies.
Paul E. McKenney4c540052010-01-14 16:10:57 -080056
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -070057When a CPU detects that it is stalling, it will print a message similar
58to the following:
59
60INFO: rcu_sched_state detected stall on CPU 5 (t=2500 jiffies)
61
62This message indicates that CPU 5 detected that it was causing a stall,
63and that the stall was affecting RCU-sched. This message will normally be
64followed by a stack dump of the offending CPU. On TREE_RCU kernel builds,
65RCU and RCU-sched are implemented by the same underlying mechanism,
66while on TREE_PREEMPT_RCU kernel builds, RCU is instead implemented
67by rcu_preempt_state.
68
69On the other hand, if the offending CPU fails to print out a stall-warning
70message quickly enough, some other CPU will print a message similar to
71the following:
72
73INFO: rcu_bh_state detected stalls on CPUs/tasks: { 3 5 } (detected by 2, 2502 jiffies)
74
75This message indicates that CPU 2 detected that CPUs 3 and 5 were both
76causing stalls, and that the stall was affecting RCU-bh. This message
77will normally be followed by stack dumps for each CPU. Please note that
78TREE_PREEMPT_RCU builds can be stalled by tasks as well as by CPUs,
79and that the tasks will be indicated by PID, for example, "P3421".
80It is even possible for a rcu_preempt_state stall to be caused by both
81CPUs -and- tasks, in which case the offending CPUs and tasks will all
82be called out in the list.
83
84Finally, if the grace period ends just as the stall warning starts
85printing, there will be a spurious stall-warning message:
86
87INFO: rcu_bh_state detected stalls on CPUs/tasks: { } (detected by 4, 2502 jiffies)
88
89This is rare, but does happen from time to time in real life.
90
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080091If the CONFIG_RCU_CPU_STALL_INFO kernel configuration parameter is set,
92more information is printed with the stall-warning message, for example:
93
94 INFO: rcu_preempt detected stall on CPU
Paul E. McKenney62310692013-03-06 13:37:09 -080095 0: (63959 ticks this GP) idle=241/3fffffffffffffff/0 softirq=82/543
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080096 (t=65000 jiffies)
97
98In kernels with CONFIG_RCU_FAST_NO_HZ, even more information is
99printed:
100
101 INFO: rcu_preempt detected stall on CPU
Paul E. McKenney62310692013-03-06 13:37:09 -0800102 0: (64628 ticks this GP) idle=dd5/3fffffffffffffff/0 softirq=82/543 last_accelerate: a345/d342 nonlazy_posted: 25 .D
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800103 (t=65000 jiffies)
104
105The "(64628 ticks this GP)" indicates that this CPU has taken more
106than 64,000 scheduling-clock interrupts during the current stalled
107grace period. If the CPU was not yet aware of the current grace
108period (for example, if it was offline), then this part of the message
109indicates how many grace periods behind the CPU is.
110
111The "idle=" portion of the message prints the dyntick-idle state.
112The hex number before the first "/" is the low-order 12 bits of the
113dynticks counter, which will have an even-numbered value if the CPU is
114in dyntick-idle mode and an odd-numbered value otherwise. The hex
115number between the two "/"s is the value of the nesting, which will
116be a small positive number if in the idle loop and a very large positive
117number (as shown above) otherwise.
118
Paul E. McKenney62310692013-03-06 13:37:09 -0800119The "softirq=" portion of the message tracks the number of RCU softirq
120handlers that the stalled CPU has executed. The number before the "/"
121is the number that had executed since boot at the time that this CPU
122last noted the beginning of a grace period, which might be the current
123(stalled) grace period, or it might be some earlier grace period (for
124example, if the CPU might have been in dyntick-idle mode for an extended
125time period. The number after the "/" is the number that have executed
126since boot until the current time. If this latter number stays constant
127across repeated stall-warning messages, it is possible that RCU's softirq
128handlers are no longer able to execute on this CPU. This can happen if
129the stalled CPU is spinning with interrupts are disabled, or, in -rt
130kernels, if a high-priority process is starving RCU's softirq handler.
131
132For CONFIG_RCU_FAST_NO_HZ kernels, the "last_accelerate:" prints the
133low-order 16 bits (in hex) of the jiffies counter when this CPU last
134invoked rcu_try_advance_all_cbs() from rcu_needs_cpu() or last invoked
135rcu_accelerate_cbs() from rcu_prepare_for_idle(). The "nonlazy_posted:"
136prints the number of non-lazy callbacks posted since the last call to
137rcu_needs_cpu(). Finally, an "L" indicates that there are currently
138no non-lazy callbacks ("." is printed otherwise, as shown above) and
139"D" indicates that dyntick-idle processing is enabled ("." is printed
140otherwise, for example, if disabled via the "nohz=" kernel boot parameter).
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800141
142
143Multiple Warnings From One Stall
144
145If a stall lasts long enough, multiple stall-warning messages will be
146printed for it. The second and subsequent messages are printed at
147longer intervals, so that the time between (say) the first and second
148message will be about three times the interval between the beginning
149of the stall and the first message.
150
151
152What Causes RCU CPU Stall Warnings?
153
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700154So your kernel printed an RCU CPU stall warning. The next question is
155"What caused it?" The following problems can result in RCU CPU stall
156warnings:
Paul E. McKenney4c540052010-01-14 16:10:57 -0800157
158o A CPU looping in an RCU read-side critical section.
159
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700160o A CPU looping with interrupts disabled. This condition can
161 result in RCU-sched and RCU-bh stalls.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800162
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700163o A CPU looping with preemption disabled. This condition can
164 result in RCU-sched stalls and, if ksoftirqd is in use, RCU-bh
165 stalls.
166
167o A CPU looping with bottom halves disabled. This condition can
168 result in RCU-sched and RCU-bh stalls.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800169
170o For !CONFIG_PREEMPT kernels, a CPU looping anywhere in the kernel
171 without invoking schedule().
172
Paul E. McKenney2c96c772010-08-23 16:34:02 -0700173o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might
174 happen to preempt a low-priority task in the middle of an RCU
175 read-side critical section. This is especially damaging if
176 that low-priority task is not permitted to run on any other CPU,
177 in which case the next RCU grace period can never complete, which
178 will eventually cause the system to run out of memory and hang.
179 While the system is in the process of running itself out of
180 memory, you might see stall-warning messages.
181
182o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that
183 is running at a higher priority than the RCU softirq threads.
184 This will prevent RCU callbacks from ever being invoked,
185 and in a CONFIG_TREE_PREEMPT_RCU kernel will further prevent
186 RCU grace periods from ever completing. Either way, the
187 system will eventually run out of memory and hang. In the
188 CONFIG_TREE_PREEMPT_RCU case, you might see stall-warning
189 messages.
190
Paul E. McKenney2c015312011-10-02 17:21:18 -0700191o A hardware or software issue shuts off the scheduler-clock
192 interrupt on a CPU that is not in dyntick-idle mode. This
193 problem really has happened, and seems to be most likely to
Frederic Weisbecker3451d022011-08-10 23:21:01 +0200194 result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels.
Paul E. McKenney2c015312011-10-02 17:21:18 -0700195
Paul E. McKenney4c540052010-01-14 16:10:57 -0800196o A bug in the RCU implementation.
197
198o A hardware failure. This is quite unlikely, but has occurred
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700199 at least once in real life. A CPU failed in a running system,
Paul E. McKenney4c540052010-01-14 16:10:57 -0800200 becoming unresponsive, but not causing an immediate crash.
201 This resulted in a series of RCU CPU stall warnings, eventually
202 leading the realization that the CPU had failed.
203
Paul E. McKenney9ceae0e2011-11-03 13:43:24 -0700204The RCU, RCU-sched, and RCU-bh implementations have CPU stall warning.
205SRCU does not have its own CPU stall warnings, but its calls to
206synchronize_sched() will result in RCU-sched detecting RCU-sched-related
207CPU stalls. Please note that RCU only detects CPU stalls when there is
208a grace period in progress. No grace period, no CPU stall warnings.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800209
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700210To diagnose the cause of the stall, inspect the stack traces.
211The offending function will usually be near the top of the stack.
212If you have a series of stall warnings from a single extended stall,
213comparing the stack traces can often help determine where the stall
214is occurring, which will usually be in the function nearest the top of
215that portion of the stack which remains the same from trace to trace.
216If you can reliably trigger the stall, ftrace can be quite helpful.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800217
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800218RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE
219and with RCU's event tracing.