blob: 90ec5341ee981a0a397710d2ffd1261eb2557896 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001RCU on Uniprocessor Systems
2
3
4A common misconception is that, on UP systems, the call_rcu() primitive
Paul E. McKenney240ebbf2009-06-25 09:08:18 -07005may immediately invoke its function. The basis of this misconception
Linus Torvalds1da177e2005-04-16 15:20:36 -07006is that since there is only one CPU, it should not be necessary to
7wait for anything else to get done, since there are no other CPUs for
Paul E. McKenneya83f1fe2005-05-01 08:59:05 -07008anything else to be happening on. Although this approach will -sort- -of-
Linus Torvalds1da177e2005-04-16 15:20:36 -07009work a surprising amount of the time, it is a very bad idea in general.
Paul E. McKenney240ebbf2009-06-25 09:08:18 -070010This document presents three examples that demonstrate exactly how bad
11an idea this is.
Linus Torvalds1da177e2005-04-16 15:20:36 -070012
13
14Example 1: softirq Suicide
15
16Suppose that an RCU-based algorithm scans a linked list containing
17elements A, B, and C in process context, and can delete elements from
18this same list in softirq context. Suppose that the process-context scan
19is referencing element B when it is interrupted by softirq processing,
20which deletes element B, and then invokes call_rcu() to free element B
21after a grace period.
22
23Now, if call_rcu() were to directly invoke its arguments, then upon return
24from softirq, the list scan would find itself referencing a newly freed
25element B. This situation can greatly decrease the life expectancy of
26your kernel.
27
Paul E. McKenneydd81eca2005-09-10 00:26:24 -070028This same problem can occur if call_rcu() is invoked from a hardware
29interrupt handler.
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32Example 2: Function-Call Fatality
33
34Of course, one could avert the suicide described in the preceding example
35by having call_rcu() directly invoke its arguments only if it was called
36from process context. However, this can fail in a similar manner.
37
38Suppose that an RCU-based algorithm again scans a linked list containing
39elements A, B, and C in process contexts, but that it invokes a function
40on each element as it is scanned. Suppose further that this function
41deletes element B from the list, then passes it to call_rcu() for deferred
42freeing. This may be a bit unconventional, but it is perfectly legal
43RCU usage, since call_rcu() must wait for a grace period to elapse.
44Therefore, in this case, allowing call_rcu() to immediately invoke
45its arguments would cause it to fail to make the fundamental guarantee
46underlying RCU, namely that call_rcu() defers invoking its arguments until
47all RCU read-side critical sections currently executing have completed.
48
Paul E. McKenneydd81eca2005-09-10 00:26:24 -070049Quick Quiz #1: why is it -not- legal to invoke synchronize_rcu() in
50 this case?
51
52
53Example 3: Death by Deadlock
54
55Suppose that call_rcu() is invoked while holding a lock, and that the
56callback function must acquire this same lock. In this case, if
57call_rcu() were to directly invoke the callback, the result would
58be self-deadlock.
59
60In some cases, it would possible to restructure to code so that
61the call_rcu() is delayed until after the lock is released. However,
62there are cases where this can be quite ugly:
63
641. If a number of items need to be passed to call_rcu() within
65 the same critical section, then the code would need to create
66 a list of them, then traverse the list once the lock was
67 released.
68
692. In some cases, the lock will be held across some kernel API,
70 so that delaying the call_rcu() until the lock is released
71 requires that the data item be passed up via a common API.
72 It is far better to guarantee that callbacks are invoked
73 with no locks held than to have to modify such APIs to allow
74 arbitrary data items to be passed back up through them.
75
76If call_rcu() directly invokes the callback, painful locking restrictions
77or API changes would be required.
78
79Quick Quiz #2: What locking restriction must RCU callbacks respect?
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81
82Summary
83
Paul E. McKenney240ebbf2009-06-25 09:08:18 -070084Permitting call_rcu() to immediately invoke its arguments breaks RCU,
85even on a UP system. So do not do it! Even on a UP system, the RCU
86infrastructure -must- respect grace periods, and -must- invoke callbacks
87from a known environment in which no locks are held.
88
89It -is- safe for synchronize_sched() and synchronize_rcu_bh() to return
90immediately on an UP system. It is also safe for synchronize_rcu()
91to return immediately on UP systems, except when running preemptable
92RCU.
93
94Quick Quiz #3: Why can't synchronize_rcu() return immediately on
95 UP systems running preemptable RCU?
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97
Paul E. McKenneydd81eca2005-09-10 00:26:24 -070098Answer to Quick Quiz #1:
99 Why is it -not- legal to invoke synchronize_rcu() in this case?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
Paul E. McKenneydd81eca2005-09-10 00:26:24 -0700101 Because the calling function is scanning an RCU-protected linked
102 list, and is therefore within an RCU read-side critical section.
103 Therefore, the called function has been invoked within an RCU
104 read-side critical section, and is not permitted to block.
105
106Answer to Quick Quiz #2:
107 What locking restriction must RCU callbacks respect?
108
109 Any lock that is acquired within an RCU callback must be
110 acquired elsewhere using an _irq variant of the spinlock
111 primitive. For example, if "mylock" is acquired by an
112 RCU callback, then a process-context acquisition of this
113 lock must use something like spin_lock_irqsave() to
114 acquire the lock.
115
116 If the process-context code were to simply use spin_lock(),
117 then, since RCU callbacks can be invoked from softirq context,
118 the callback might be called from a softirq that interrupted
119 the process-context critical section. This would result in
120 self-deadlock.
121
122 This restriction might seem gratuitous, since very few RCU
123 callbacks acquire locks directly. However, a great many RCU
124 callbacks do acquire locks -indirectly-, for example, via
125 the kfree() primitive.
Paul E. McKenney240ebbf2009-06-25 09:08:18 -0700126
127Answer to Quick Quiz #3:
128 Why can't synchronize_rcu() return immediately on UP systems
129 running preemptable RCU?
130
131 Because some other task might have been preempted in the middle
132 of an RCU read-side critical section. If synchronize_rcu()
133 simply immediately returned, it would prematurely signal the
134 end of the grace period, which would come as a nasty shock to
135 that other thread when it started running again.