blob: 0f7fb4298e7e0b657a7df678ee0852de7ab7887d [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. McKenney64d3b7a2013-08-19 11:59:43 -070015 21 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
Xie XiuQi84596cc2014-11-11 12:03:26 +080018 /sys/module/rcupdate/parameters/rcu_cpu_stall_timeout, however
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080019 this parameter is checked only at the beginning of a cycle.
Paul E. McKenney64d3b7a2013-08-19 11:59:43 -070020 So if you are 10 seconds into a 40-second stall, setting this
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080021 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
Paul E. McKenney96224da2014-02-25 09:47:34 -080027 /sys/module/rcupdate/parameters/rcu_cpu_stall_suppress.
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080028
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080029RCU_STALL_DELAY_DELTA
30
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
Paul E. McKenney64d3b7a2013-08-19 11:59:43 -070034 giving an RCU CPU stall warning message. (This is a cpp
35 macro, not a kernel configuration parameter.)
Paul E. McKenney4c540052010-01-14 16:10:57 -080036
37RCU_STALL_RAT_DELAY
38
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -070039 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
Paul E. McKenney64d3b7a2013-08-19 11:59:43 -070044 two jiffies. (This is a cpp macro, not a kernel configuration
45 parameter.)
Paul E. McKenney4c540052010-01-14 16:10:57 -080046
Paul E. McKenney37fe5f02014-07-29 09:49:23 -070047rcupdate.rcu_task_stall_timeout
48
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:
53
54 INFO: rcu_tasks detected stalls on tasks:
55
56 And continues with the output of sched_show_task() for each
57 task stalling the current RCU-tasks grace period.
58
59For non-RCU-tasks flavors of RCU, when a CPU detects that it is stalling,
60it will print a message similar to the following:
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -070061
62INFO: rcu_sched_state detected stall on CPU 5 (t=2500 jiffies)
63
64This message indicates that CPU 5 detected that it was causing a stall,
65and that the stall was affecting RCU-sched. This message will normally be
66followed by a stack dump of the offending CPU. On TREE_RCU kernel builds,
67RCU and RCU-sched are implemented by the same underlying mechanism,
Pranith Kumar28f65692014-09-22 14:00:48 -040068while on PREEMPT_RCU kernel builds, RCU is instead implemented
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -070069by rcu_preempt_state.
70
71On the other hand, if the offending CPU fails to print out a stall-warning
72message quickly enough, some other CPU will print a message similar to
73the following:
74
75INFO: rcu_bh_state detected stalls on CPUs/tasks: { 3 5 } (detected by 2, 2502 jiffies)
76
77This message indicates that CPU 2 detected that CPUs 3 and 5 were both
78causing stalls, and that the stall was affecting RCU-bh. This message
79will normally be followed by stack dumps for each CPU. Please note that
Pranith Kumar28f65692014-09-22 14:00:48 -040080PREEMPT_RCU builds can be stalled by tasks as well as by CPUs,
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -070081and that the tasks will be indicated by PID, for example, "P3421".
82It is even possible for a rcu_preempt_state stall to be caused by both
83CPUs -and- tasks, in which case the offending CPUs and tasks will all
84be called out in the list.
85
86Finally, if the grace period ends just as the stall warning starts
87printing, there will be a spurious stall-warning message:
88
89INFO: rcu_bh_state detected stalls on CPUs/tasks: { } (detected by 4, 2502 jiffies)
90
Paul E. McKenney64d3b7a2013-08-19 11:59:43 -070091This is rare, but does happen from time to time in real life. It is also
92possible for a zero-jiffy stall to be flagged in this case, depending
93on how the stall warning and the grace-period initialization happen to
94interact. Please note that it is not possible to entirely eliminate this
95sort of false positive without resorting to things like stop_machine(),
96which is overkill for this sort of problem.
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -070097
Paul E. McKenney75c27f12015-06-11 15:22:43 -070098Recent kernels will print a long form of the stall-warning message:
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -080099
100 INFO: rcu_preempt detected stall on CPU
Paul E. McKenney62310692013-03-06 13:37:09 -0800101 0: (63959 ticks this GP) idle=241/3fffffffffffffff/0 softirq=82/543
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800102 (t=65000 jiffies)
103
Paul E. McKenney75c27f12015-06-11 15:22:43 -0700104In kernels with CONFIG_RCU_FAST_NO_HZ, more information is printed:
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800105
106 INFO: rcu_preempt detected stall on CPU
Paul E. McKenney62310692013-03-06 13:37:09 -0800107 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 -0800108 (t=65000 jiffies)
109
110The "(64628 ticks this GP)" indicates that this CPU has taken more
111than 64,000 scheduling-clock interrupts during the current stalled
112grace period. If the CPU was not yet aware of the current grace
113period (for example, if it was offline), then this part of the message
114indicates how many grace periods behind the CPU is.
115
116The "idle=" portion of the message prints the dyntick-idle state.
117The hex number before the first "/" is the low-order 12 bits of the
118dynticks counter, which will have an even-numbered value if the CPU is
119in dyntick-idle mode and an odd-numbered value otherwise. The hex
120number between the two "/"s is the value of the nesting, which will
121be a small positive number if in the idle loop and a very large positive
122number (as shown above) otherwise.
123
Paul E. McKenney62310692013-03-06 13:37:09 -0800124The "softirq=" portion of the message tracks the number of RCU softirq
125handlers that the stalled CPU has executed. The number before the "/"
126is the number that had executed since boot at the time that this CPU
127last 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
129example, if the CPU might have been in dyntick-idle mode for an extended
130time period. The number after the "/" is the number that have executed
131since boot until the current time. If this latter number stays constant
132across repeated stall-warning messages, it is possible that RCU's softirq
133handlers are no longer able to execute on this CPU. This can happen if
134the stalled CPU is spinning with interrupts are disabled, or, in -rt
135kernels, if a high-priority process is starving RCU's softirq handler.
136
137For CONFIG_RCU_FAST_NO_HZ kernels, the "last_accelerate:" prints the
138low-order 16 bits (in hex) of the jiffies counter when this CPU last
139invoked rcu_try_advance_all_cbs() from rcu_needs_cpu() or last invoked
140rcu_accelerate_cbs() from rcu_prepare_for_idle(). The "nonlazy_posted:"
141prints the number of non-lazy callbacks posted since the last call to
142rcu_needs_cpu(). Finally, an "L" indicates that there are currently
143no non-lazy callbacks ("." is printed otherwise, as shown above) and
144"D" indicates that dyntick-idle processing is enabled ("." is printed
145otherwise, for example, if disabled via the "nohz=" kernel boot parameter).
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800146
Paul E. McKenneyfb81a442014-12-17 08:35:02 -0800147If the relevant grace-period kthread has been unable to run prior to
148the stall warning, the following additional line is printed:
149
150 rcu_preempt kthread starved for 2023 jiffies!
151
152Starving the grace-period kthreads of CPU time can of course result in
153RCU CPU stall warnings even when all CPUs and tasks have passed through
154the required quiescent states.
155
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800156
157Multiple Warnings From One Stall
158
159If a stall lasts long enough, multiple stall-warning messages will be
160printed for it. The second and subsequent messages are printed at
161longer intervals, so that the time between (say) the first and second
162message will be about three times the interval between the beginning
163of the stall and the first message.
164
165
Paul E. McKenney99a930b2015-06-30 14:54:09 -0700166Stall Warnings for Expedited Grace Periods
167
168If an expedited grace period detects a stall, it will place a message
169like the following in dmesg:
170
171 INFO: rcu_sched detected expedited stalls on CPUs: { 1 2 6 } 26009 jiffies s: 1043
172
173This indicates that CPUs 1, 2, and 6 have failed to respond to a
174reschedule IPI, that the expedited grace period has been going on for
17526,009 jiffies, and that the expedited grace-period sequence counter is
1761043. The fact that this last value is odd indicates that an expedited
177grace period is in flight.
178
179It is entirely possible to see stall warnings from normal and from
180expedited grace periods at about the same time from the same run.
181
182
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800183What Causes RCU CPU Stall Warnings?
184
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700185So 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
187warnings:
Paul E. McKenney4c540052010-01-14 16:10:57 -0800188
189o A CPU looping in an RCU read-side critical section.
190
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700191o A CPU looping with interrupts disabled. This condition can
192 result in RCU-sched and RCU-bh stalls.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800193
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700194o A CPU looping with preemption disabled. This condition can
195 result in RCU-sched stalls and, if ksoftirqd is in use, RCU-bh
196 stalls.
197
198o A CPU looping with bottom halves disabled. This condition can
199 result in RCU-sched and RCU-bh stalls.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800200
Paul E. McKenney37fe5f02014-07-29 09:49:23 -0700201o 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().
Paul E. McKenney4c540052010-01-14 16:10:57 -0800207
Paul E. McKenneyda873de2015-08-04 11:54:04 -0700208o Booting Linux using a console connection that is too slow to
209 keep up with the boot-time console-message rate. For example,
210 a 115Kbaud serial console can be -way- too slow to keep up
211 with boot-time message rates, and will frequently result in
212 RCU CPU stall warning messages. Especially if you have added
213 debug printk()s.
214
Paul E. McKenney6ccd2ec2014-12-11 10:20:59 -0800215o Anything that prevents RCU's grace-period kthreads from running.
216 This can result in the "All QSes seen" console-log message.
217 This message will include information on when the kthread last
218 ran and how often it should be expected to run.
219
Paul E. McKenney2c96c772010-08-23 16:34:02 -0700220o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might
221 happen to preempt a low-priority task in the middle of an RCU
222 read-side critical section. This is especially damaging if
223 that low-priority task is not permitted to run on any other CPU,
224 in which case the next RCU grace period can never complete, which
225 will eventually cause the system to run out of memory and hang.
226 While the system is in the process of running itself out of
227 memory, you might see stall-warning messages.
228
229o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that
230 is running at a higher priority than the RCU softirq threads.
231 This will prevent RCU callbacks from ever being invoked,
Pranith Kumar28f65692014-09-22 14:00:48 -0400232 and in a CONFIG_PREEMPT_RCU kernel will further prevent
Paul E. McKenney2c96c772010-08-23 16:34:02 -0700233 RCU grace periods from ever completing. Either way, the
234 system will eventually run out of memory and hang. In the
Pranith Kumar28f65692014-09-22 14:00:48 -0400235 CONFIG_PREEMPT_RCU case, you might see stall-warning
Paul E. McKenney2c96c772010-08-23 16:34:02 -0700236 messages.
237
Paul E. McKenney2c015312011-10-02 17:21:18 -0700238o A hardware or software issue shuts off the scheduler-clock
239 interrupt on a CPU that is not in dyntick-idle mode. This
240 problem really has happened, and seems to be most likely to
Frederic Weisbecker3451d022011-08-10 23:21:01 +0200241 result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels.
Paul E. McKenney2c015312011-10-02 17:21:18 -0700242
Paul E. McKenney4c540052010-01-14 16:10:57 -0800243o A bug in the RCU implementation.
244
245o A hardware failure. This is quite unlikely, but has occurred
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700246 at least once in real life. A CPU failed in a running system,
Paul E. McKenney4c540052010-01-14 16:10:57 -0800247 becoming unresponsive, but not causing an immediate crash.
248 This resulted in a series of RCU CPU stall warnings, eventually
249 leading the realization that the CPU had failed.
250
Paul E. McKenney37fe5f02014-07-29 09:49:23 -0700251The RCU, RCU-sched, RCU-bh, and RCU-tasks implementations have CPU stall
252warning. Note that SRCU does -not- have CPU stall warnings. Please note
253that RCU only detects CPU stalls when there is a grace period in progress.
254No grace period, no CPU stall warnings.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800255
Paul E. McKenneyf1d507b2010-04-15 15:49:46 -0700256To diagnose the cause of the stall, inspect the stack traces.
257The offending function will usually be near the top of the stack.
258If you have a series of stall warnings from a single extended stall,
259comparing the stack traces can often help determine where the stall
260is occurring, which will usually be in the function nearest the top of
261that portion of the stack which remains the same from trace to trace.
262If you can reliably trigger the stall, ftrace can be quite helpful.
Paul E. McKenney4c540052010-01-14 16:10:57 -0800263
Paul E. McKenney24cd7fd2012-01-20 17:35:55 -0800264RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE
Paul E. McKenney64d3b7a2013-08-19 11:59:43 -0700265and with RCU's event tracing. For information on RCU's event tracing,
266see include/trace/events/rcu.h.