1Using RCU's CPU Stall Detector 2 3The 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: 9 10CONFIG_RCU_CPU_STALL_TIMEOUT 11 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 15 21 seconds. 16 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. 25 26 Stall-warning messages may be enabled and disabled completely via 27 /sys/module/rcupdate/parameters/rcu_cpu_stall_suppress. 28 29RCU_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 34 giving an RCU CPU stall warning message. (This is a cpp 35 macro, not a kernel configuration parameter.) 36 37RCU_STALL_RAT_DELAY 38 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 45 parameter.) 46 47rcupdate.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: 61 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, 68while on PREEMPT_RCU kernel builds, RCU is instead implemented 69by 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 80PREEMPT_RCU builds can be stalled by tasks as well as by CPUs, 81and 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 91This 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. 97 98Recent kernels will print a long form of the stall-warning message: 99 100 INFO: rcu_preempt detected stall on CPU 101 0: (63959 ticks this GP) idle=241/3fffffffffffffff/0 softirq=82/543 102 (t=65000 jiffies) 103 104In kernels with CONFIG_RCU_FAST_NO_HZ, more information is printed: 105 106 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 108 (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 124The "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). 146 147If 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. 155 156 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 166Stall 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 183What Causes RCU CPU Stall Warnings? 184 185So 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: 188 189o A CPU looping in an RCU read-side critical section. 190 191o A CPU looping with interrupts disabled. This condition can 192 result in RCU-sched and RCU-bh stalls. 193 194o 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. 200 201o 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(). 207 208o 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 215o 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 220o 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, 232 and in a CONFIG_PREEMPT_RCU kernel will further prevent 233 RCU grace periods from ever completing. Either way, the 234 system will eventually run out of memory and hang. In the 235 CONFIG_PREEMPT_RCU case, you might see stall-warning 236 messages. 237 238o 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 241 result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels. 242 243o A bug in the RCU implementation. 244 245o A hardware failure. This is quite unlikely, but has occurred 246 at least once in real life. A CPU failed in a running system, 247 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 251The 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. 255 256To 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. 263 264RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE 265and with RCU's event tracing. For information on RCU's event tracing, 266see include/trace/events/rcu.h. 267