UP.rst (5978B)
1.. _up_doc: 2 3RCU on Uniprocessor Systems 4=========================== 5 6A common misconception is that, on UP systems, the call_rcu() primitive 7may immediately invoke its function. The basis of this misconception 8is that since there is only one CPU, it should not be necessary to 9wait for anything else to get done, since there are no other CPUs for 10anything else to be happening on. Although this approach will *sort of* 11work a surprising amount of the time, it is a very bad idea in general. 12This document presents three examples that demonstrate exactly how bad 13an idea this is. 14 15Example 1: softirq Suicide 16-------------------------- 17 18Suppose that an RCU-based algorithm scans a linked list containing 19elements A, B, and C in process context, and can delete elements from 20this same list in softirq context. Suppose that the process-context scan 21is referencing element B when it is interrupted by softirq processing, 22which deletes element B, and then invokes call_rcu() to free element B 23after a grace period. 24 25Now, if call_rcu() were to directly invoke its arguments, then upon return 26from softirq, the list scan would find itself referencing a newly freed 27element B. This situation can greatly decrease the life expectancy of 28your kernel. 29 30This same problem can occur if call_rcu() is invoked from a hardware 31interrupt handler. 32 33Example 2: Function-Call Fatality 34--------------------------------- 35 36Of course, one could avert the suicide described in the preceding example 37by having call_rcu() directly invoke its arguments only if it was called 38from process context. However, this can fail in a similar manner. 39 40Suppose that an RCU-based algorithm again scans a linked list containing 41elements A, B, and C in process contexts, but that it invokes a function 42on each element as it is scanned. Suppose further that this function 43deletes element B from the list, then passes it to call_rcu() for deferred 44freeing. This may be a bit unconventional, but it is perfectly legal 45RCU usage, since call_rcu() must wait for a grace period to elapse. 46Therefore, in this case, allowing call_rcu() to immediately invoke 47its arguments would cause it to fail to make the fundamental guarantee 48underlying RCU, namely that call_rcu() defers invoking its arguments until 49all RCU read-side critical sections currently executing have completed. 50 51Quick Quiz #1: 52 Why is it *not* legal to invoke synchronize_rcu() in this case? 53 54:ref:`Answers to Quick Quiz <answer_quick_quiz_up>` 55 56Example 3: Death by Deadlock 57---------------------------- 58 59Suppose that call_rcu() is invoked while holding a lock, and that the 60callback function must acquire this same lock. In this case, if 61call_rcu() were to directly invoke the callback, the result would 62be self-deadlock. 63 64In some cases, it would possible to restructure to code so that 65the call_rcu() is delayed until after the lock is released. However, 66there are cases where this can be quite ugly: 67 681. If a number of items need to be passed to call_rcu() within 69 the same critical section, then the code would need to create 70 a list of them, then traverse the list once the lock was 71 released. 72 732. In some cases, the lock will be held across some kernel API, 74 so that delaying the call_rcu() until the lock is released 75 requires that the data item be passed up via a common API. 76 It is far better to guarantee that callbacks are invoked 77 with no locks held than to have to modify such APIs to allow 78 arbitrary data items to be passed back up through them. 79 80If call_rcu() directly invokes the callback, painful locking restrictions 81or API changes would be required. 82 83Quick Quiz #2: 84 What locking restriction must RCU callbacks respect? 85 86:ref:`Answers to Quick Quiz <answer_quick_quiz_up>` 87 88Summary 89------- 90 91Permitting call_rcu() to immediately invoke its arguments breaks RCU, 92even on a UP system. So do not do it! Even on a UP system, the RCU 93infrastructure *must* respect grace periods, and *must* invoke callbacks 94from a known environment in which no locks are held. 95 96Note that it *is* safe for synchronize_rcu() to return immediately on 97UP systems, including PREEMPT SMP builds running on UP systems. 98 99Quick Quiz #3: 100 Why can't synchronize_rcu() return immediately on UP systems running 101 preemptable RCU? 102 103.. _answer_quick_quiz_up: 104 105Answer to Quick Quiz #1: 106 Why is it *not* legal to invoke synchronize_rcu() in this case? 107 108 Because the calling function is scanning an RCU-protected linked 109 list, and is therefore within an RCU read-side critical section. 110 Therefore, the called function has been invoked within an RCU 111 read-side critical section, and is not permitted to block. 112 113Answer to Quick Quiz #2: 114 What locking restriction must RCU callbacks respect? 115 116 Any lock that is acquired within an RCU callback must be acquired 117 elsewhere using an _bh variant of the spinlock primitive. 118 For example, if "mylock" is acquired by an RCU callback, then 119 a process-context acquisition of this lock must use something 120 like spin_lock_bh() to acquire the lock. Please note that 121 it is also OK to use _irq variants of spinlocks, for example, 122 spin_lock_irqsave(). 123 124 If the process-context code were to simply use spin_lock(), 125 then, since RCU callbacks can be invoked from softirq context, 126 the callback might be called from a softirq that interrupted 127 the process-context critical section. This would result in 128 self-deadlock. 129 130 This restriction might seem gratuitous, since very few RCU 131 callbacks acquire locks directly. However, a great many RCU 132 callbacks do acquire locks *indirectly*, for example, via 133 the kfree() primitive. 134 135Answer to Quick Quiz #3: 136 Why can't synchronize_rcu() return immediately on UP systems 137 running preemptable RCU? 138 139 Because some other task might have been preempted in the middle 140 of an RCU read-side critical section. If synchronize_rcu() 141 simply immediately returned, it would prematurely signal the 142 end of the grace period, which would come as a nasty shock to 143 that other thread when it started running again.