1Using RCU's CPU Stall Detector 2 3This document first discusses what sorts of issues RCU's CPU stall 4detector can locate, and then discusses kernel parameters and Kconfig 5options that can be used to fine-tune the detector's operation. Finally, 6this document explains the stall detector's "splat" format. 7 8 9What Causes RCU CPU Stall Warnings? 10 11So your kernel printed an RCU CPU stall warning. The next question is 12"What caused it?" The following problems can result in RCU CPU stall 13warnings: 14 15o A CPU looping in an RCU read-side critical section. 16 17o A CPU looping with interrupts disabled. 18 19o A CPU looping with preemption disabled. This condition can 20 result in RCU-sched stalls and, if ksoftirqd is in use, RCU-bh 21 stalls. 22 23o A CPU looping with bottom halves disabled. This condition can 24 result in RCU-sched and RCU-bh stalls. 25 26o For !CONFIG_PREEMPT kernels, a CPU looping anywhere in the 27 kernel without invoking schedule(). Note that cond_resched() 28 does not necessarily prevent RCU CPU stall warnings. Therefore, 29 if the looping in the kernel is really expected and desirable 30 behavior, you might need to replace some of the cond_resched() 31 calls with calls to cond_resched_rcu_qs(). 32 33o Booting Linux using a console connection that is too slow to 34 keep up with the boot-time console-message rate. For example, 35 a 115Kbaud serial console can be -way- too slow to keep up 36 with boot-time message rates, and will frequently result in 37 RCU CPU stall warning messages. Especially if you have added 38 debug printk()s. 39 40o Anything that prevents RCU's grace-period kthreads from running. 41 This can result in the "All QSes seen" console-log message. 42 This message will include information on when the kthread last 43 ran and how often it should be expected to run. 44 45o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might 46 happen to preempt a low-priority task in the middle of an RCU 47 read-side critical section. This is especially damaging if 48 that low-priority task is not permitted to run on any other CPU, 49 in which case the next RCU grace period can never complete, which 50 will eventually cause the system to run out of memory and hang. 51 While the system is in the process of running itself out of 52 memory, you might see stall-warning messages. 53 54o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that 55 is running at a higher priority than the RCU softirq threads. 56 This will prevent RCU callbacks from ever being invoked, 57 and in a CONFIG_PREEMPT_RCU kernel will further prevent 58 RCU grace periods from ever completing. Either way, the 59 system will eventually run out of memory and hang. In the 60 CONFIG_PREEMPT_RCU case, you might see stall-warning 61 messages. 62 63o A hardware or software issue shuts off the scheduler-clock 64 interrupt on a CPU that is not in dyntick-idle mode. This 65 problem really has happened, and seems to be most likely to 66 result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels. 67 68o A bug in the RCU implementation. 69 70o A hardware failure. This is quite unlikely, but has occurred 71 at least once in real life. A CPU failed in a running system, 72 becoming unresponsive, but not causing an immediate crash. 73 This resulted in a series of RCU CPU stall warnings, eventually 74 leading the realization that the CPU had failed. 75 76The RCU, RCU-sched, RCU-bh, and RCU-tasks implementations have CPU stall 77warning. Note that SRCU does -not- have CPU stall warnings. Please note 78that RCU only detects CPU stalls when there is a grace period in progress. 79No grace period, no CPU stall warnings. 80 81To diagnose the cause of the stall, inspect the stack traces. 82The offending function will usually be near the top of the stack. 83If you have a series of stall warnings from a single extended stall, 84comparing the stack traces can often help determine where the stall 85is occurring, which will usually be in the function nearest the top of 86that portion of the stack which remains the same from trace to trace. 87If you can reliably trigger the stall, ftrace can be quite helpful. 88 89RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE 90and with RCU's event tracing. For information on RCU's event tracing, 91see include/trace/events/rcu.h. 92 93 94Fine-Tuning the RCU CPU Stall Detector 95 96The rcuupdate.rcu_cpu_stall_suppress module parameter disables RCU's 97CPU stall detector, which detects conditions that unduly delay RCU grace 98periods. This module parameter enables CPU stall detection by default, 99but may be overridden via boot-time parameter or at runtime via sysfs. 100The stall detector's idea of what constitutes "unduly delayed" is 101controlled by a set of kernel configuration variables and cpp macros: 102 103CONFIG_RCU_CPU_STALL_TIMEOUT 104 105 This kernel configuration parameter defines the period of time 106 that RCU will wait from the beginning of a grace period until it 107 issues an RCU CPU stall warning. This time period is normally 108 21 seconds. 109 110 This configuration parameter may be changed at runtime via the 111 /sys/module/rcupdate/parameters/rcu_cpu_stall_timeout, however 112 this parameter is checked only at the beginning of a cycle. 113 So if you are 10 seconds into a 40-second stall, setting this 114 sysfs parameter to (say) five will shorten the timeout for the 115 -next- stall, or the following warning for the current stall 116 (assuming the stall lasts long enough). It will not affect the 117 timing of the next warning for the current stall. 118 119 Stall-warning messages may be enabled and disabled completely via 120 /sys/module/rcupdate/parameters/rcu_cpu_stall_suppress. 121 122RCU_STALL_DELAY_DELTA 123 124 Although the lockdep facility is extremely useful, it does add 125 some overhead. Therefore, under CONFIG_PROVE_RCU, the 126 RCU_STALL_DELAY_DELTA macro allows five extra seconds before 127 giving an RCU CPU stall warning message. (This is a cpp 128 macro, not a kernel configuration parameter.) 129 130RCU_STALL_RAT_DELAY 131 132 The CPU stall detector tries to make the offending CPU print its 133 own warnings, as this often gives better-quality stack traces. 134 However, if the offending CPU does not detect its own stall in 135 the number of jiffies specified by RCU_STALL_RAT_DELAY, then 136 some other CPU will complain. This delay is normally set to 137 two jiffies. (This is a cpp macro, not a kernel configuration 138 parameter.) 139 140rcupdate.rcu_task_stall_timeout 141 142 This boot/sysfs parameter controls the RCU-tasks stall warning 143 interval. A value of zero or less suppresses RCU-tasks stall 144 warnings. A positive value sets the stall-warning interval 145 in jiffies. An RCU-tasks stall warning starts with the line: 146 147 INFO: rcu_tasks detected stalls on tasks: 148 149 And continues with the output of sched_show_task() for each 150 task stalling the current RCU-tasks grace period. 151 152 153Interpreting RCU's CPU Stall-Detector "Splats" 154 155For non-RCU-tasks flavors of RCU, when a CPU detects that it is stalling, 156it will print a message similar to the following: 157 158INFO: rcu_sched_state detected stall on CPU 5 (t=2500 jiffies) 159 160This message indicates that CPU 5 detected that it was causing a stall, 161and that the stall was affecting RCU-sched. This message will normally be 162followed by a stack dump of the offending CPU. On TREE_RCU kernel builds, 163RCU and RCU-sched are implemented by the same underlying mechanism, 164while on PREEMPT_RCU kernel builds, RCU is instead implemented 165by rcu_preempt_state. 166 167On the other hand, if the offending CPU fails to print out a stall-warning 168message quickly enough, some other CPU will print a message similar to 169the following: 170 171INFO: rcu_bh_state detected stalls on CPUs/tasks: { 3 5 } (detected by 2, 2502 jiffies) 172 173This message indicates that CPU 2 detected that CPUs 3 and 5 were both 174causing stalls, and that the stall was affecting RCU-bh. This message 175will normally be followed by stack dumps for each CPU. Please note that 176PREEMPT_RCU builds can be stalled by tasks as well as by CPUs, 177and that the tasks will be indicated by PID, for example, "P3421". 178It is even possible for a rcu_preempt_state stall to be caused by both 179CPUs -and- tasks, in which case the offending CPUs and tasks will all 180be called out in the list. 181 182Finally, if the grace period ends just as the stall warning starts 183printing, there will be a spurious stall-warning message: 184 185INFO: rcu_bh_state detected stalls on CPUs/tasks: { } (detected by 4, 2502 jiffies) 186 187This is rare, but does happen from time to time in real life. It is also 188possible for a zero-jiffy stall to be flagged in this case, depending 189on how the stall warning and the grace-period initialization happen to 190interact. Please note that it is not possible to entirely eliminate this 191sort of false positive without resorting to things like stop_machine(), 192which is overkill for this sort of problem. 193 194Recent kernels will print a long form of the stall-warning message: 195 196 INFO: rcu_preempt detected stall on CPU 197 0: (63959 ticks this GP) idle=241/3fffffffffffffff/0 softirq=82/543 198 (t=65000 jiffies) 199 200In kernels with CONFIG_RCU_FAST_NO_HZ, more information is printed: 201 202 INFO: rcu_preempt detected stall on CPU 203 0: (64628 ticks this GP) idle=dd5/3fffffffffffffff/0 softirq=82/543 last_accelerate: a345/d342 nonlazy_posted: 25 .D 204 (t=65000 jiffies) 205 206The "(64628 ticks this GP)" indicates that this CPU has taken more 207than 64,000 scheduling-clock interrupts during the current stalled 208grace period. If the CPU was not yet aware of the current grace 209period (for example, if it was offline), then this part of the message 210indicates how many grace periods behind the CPU is. 211 212The "idle=" portion of the message prints the dyntick-idle state. 213The hex number before the first "/" is the low-order 12 bits of the 214dynticks counter, which will have an even-numbered value if the CPU is 215in dyntick-idle mode and an odd-numbered value otherwise. The hex 216number between the two "/"s is the value of the nesting, which will 217be a small positive number if in the idle loop and a very large positive 218number (as shown above) otherwise. 219 220The "softirq=" portion of the message tracks the number of RCU softirq 221handlers that the stalled CPU has executed. The number before the "/" 222is the number that had executed since boot at the time that this CPU 223last noted the beginning of a grace period, which might be the current 224(stalled) grace period, or it might be some earlier grace period (for 225example, if the CPU might have been in dyntick-idle mode for an extended 226time period. The number after the "/" is the number that have executed 227since boot until the current time. If this latter number stays constant 228across repeated stall-warning messages, it is possible that RCU's softirq 229handlers are no longer able to execute on this CPU. This can happen if 230the stalled CPU is spinning with interrupts are disabled, or, in -rt 231kernels, if a high-priority process is starving RCU's softirq handler. 232 233For CONFIG_RCU_FAST_NO_HZ kernels, the "last_accelerate:" prints the 234low-order 16 bits (in hex) of the jiffies counter when this CPU last 235invoked rcu_try_advance_all_cbs() from rcu_needs_cpu() or last invoked 236rcu_accelerate_cbs() from rcu_prepare_for_idle(). The "nonlazy_posted:" 237prints the number of non-lazy callbacks posted since the last call to 238rcu_needs_cpu(). Finally, an "L" indicates that there are currently 239no non-lazy callbacks ("." is printed otherwise, as shown above) and 240"D" indicates that dyntick-idle processing is enabled ("." is printed 241otherwise, for example, if disabled via the "nohz=" kernel boot parameter). 242 243If the relevant grace-period kthread has been unable to run prior to 244the stall warning, the following additional line is printed: 245 246 rcu_preempt kthread starved for 2023 jiffies! 247 248Starving the grace-period kthreads of CPU time can of course result in 249RCU CPU stall warnings even when all CPUs and tasks have passed through 250the required quiescent states. 251 252 253Multiple Warnings From One Stall 254 255If a stall lasts long enough, multiple stall-warning messages will be 256printed for it. The second and subsequent messages are printed at 257longer intervals, so that the time between (say) the first and second 258message will be about three times the interval between the beginning 259of the stall and the first message. 260 261 262Stall Warnings for Expedited Grace Periods 263 264If an expedited grace period detects a stall, it will place a message 265like the following in dmesg: 266 267 INFO: rcu_sched detected expedited stalls on CPUs: { 1 2 6 } 26009 jiffies s: 1043 268 269This indicates that CPUs 1, 2, and 6 have failed to respond to a 270reschedule IPI, that the expedited grace period has been going on for 27126,009 jiffies, and that the expedited grace-period sequence counter is 2721043. The fact that this last value is odd indicates that an expedited 273grace period is in flight. 274 275It is entirely possible to see stall warnings from normal and from 276expedited grace periods at about the same time from the same run. 277